Dynamics of Civil Structures, Volume 2 Proceedings of the 35th IMAC, A Conference and Exposition on Structural Dynamics 2017 [1st edition 2017] 9783319547770, 9783319547763, 3319547763, 3319547771

Citation preview

Conference Proceedings of the Society for Experimental Mechanics Series

Juan Caicedo · Shamim Pakzad  Editors

Dynamics of Civil Structures, Volume 2 Proceedings of the 35th IMAC, A Conference and Exposition on Structural Dynamics 2017

Conference Proceedings of the Society for Experimental Mechanics Series

Series Editor Kristin B. Zimmerman, Ph.D. Society for Experimental Mechanics, Inc., Bethel, CT, USA

More information about this series at http://www.springer.com/series/8922

Juan Caicedo • Shamim Pakzad Editors

Dynamics of Civil Structures, Volume 2 Proceedings of the 35th IMAC, A Conference and Exposition on Structural Dynamics 2017

123

Editors Juan Caicedo Department of Civil and Environmental Engineering University of South Carolina Columbia, SC, USA

Shamim Pakzad Department of Civil and Environmental Engineering Lehigh University Bethlehem, PA, USA

ISSN 2191-5644 ISSN 2191-5652 (electronic) Conference Proceedings of the Society for Experimental Mechanics Series ISBN 978-3-319-54776-3 ISBN 978-3-319-54777-0 (eBook) DOI 10.1007/978-3-319-54777-0 Library of Congress Control Number: 2017938029 © The Society for Experimental Mechanics, Inc. 2017 This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. The publisher, the authors and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, express or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Printed on acid-free paper This Springer imprint is published by Springer Nature The registered company is Springer International Publishing AG The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland

Preface

Dynamics of Civil Structures represents one of ten volumes of technical papers presented at the 35th IMAC, A Conference and Exposition on Structural Dynamics, organized by the Society for Experimental Mechanics, and held in Garden Grove, California, January 30–February 2, 2017. The full proceedings also include volumes on Nonlinear Dynamics; Model Validation and Uncertainty Quantification; Dynamics of Coupled Structures; Sensors and Instrumentation; Special Topics in Structural Dynamics; Structural Health Monitoring and Damage Detection; Rotating Machinery, Hybrid Test Methods, Vibro-Acoustics & Laser Vibrometry; Shock & Vibration, Aircraft/Aerospace and Energy Harvesting; and Topics in Modal Analysis & Testing. Each collection presents early findings from analytical, experimental, and computational investigations on an important area within structural dynamics. Dynamics of Civil Structures is one of these areas which cover topics of interest of several disciplines in engineering and science. The Dynamics of Civil Structures Technical Division serves as a primary focal point within the SEM umbrella for technical activities devoted to civil structure analysis, testing, monitoring, and assessment. This volume covers a variety of topics including damage identification, human-structure interaction, hybrid testing, vibration control, model updating, modal analysis of in-service structures, sensing and measurements of structural systems, and bridge dynamics. Papers cover testing and analysis of all kinds of civil engineering structures such as buildings, bridges, stadiums, dams, and others. The organizers would like to thank the authors, presenters, session organizers, and session chairs for their participation in this track. Columbia, SC, USA Bethlehem, PA, USA

Juan Caicedo Shamim Pakzad

v

Contents

1

2

3

Semi-Active Base Isolation of Civil Engineering Structures Based on Optimal Viscous Damping and Zero Dynamic Stiffness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Felix Weber, Hans Distl, and Christian Braun

1

Long-Term Performance of Specialized Fluid Dampers Under Continuous Vibration on a Pedestrian Bridge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Alan R. Klembczyk

11

Analysis of Variation Rate of Displacement to Temperature of Service Stage Cable-Stayed Bridge Using Temperatures and Displacement Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Hyun-Joong Kim

21

4

Triple Friction Pendulum: Does It Improve the Isolation Performance? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Felix Weber, Peter Huber, Hans Distl, and Christian Braun

27

5

Experimental Investigation of the Dynamic Characteristics of a Glass-FRP Suspension Footbridge . . . . . . . . . Xiaojun Wei, Justin Russell, Stana Živanovi´c, and J. Toby Mottram

37

6

Vibration-Based Occupant Detection Using a Multiple-Model Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Yves Reuland, Sai G. S. Pai, Slah Drira, and Ian F. C. Smith

49

7

Vibration Assessment and Control in Technical Facilities Using an Integrated Multidisciplinary Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Nicholas Christie, James Hargreaves, Rob Harrison, and Francois Lancelot

57

8

Iterative Pole-Zero Model Updating Using Multiple Frequency Response Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . M. Dorosti, R.H.B. Fey, M.F. Heertjes, M.M.J. van de Wal, and H. Nijmeijer

65

9

Vision-Based Concrete Crack Detection Using a Convolutional Neural Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Young-Jin Cha and Wooram Choi

71

10

Analytical and Experimental Analysis of Rocking Columns Subject to Seismic Excitation . . . . . . . . . . . . . . . . . . . . Ryan Kent Giles and Thomas John Kennedy

75

11

Extending the Fixed-Points Technique for Optimum Design of Rotational Inertial Tuned Mass Dampers . . . Abdollah Javidialesaadi and Nicholas Wierschem

83

12

Temperature Effects on the Modal Properties of a Suspension Bridge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Etienne Cheynet, Jonas Snæbjörnsson, and Jasna Bogunovi´c Jakobsen

87

13

Mass Scaling of Mode Shapes Based on the Effect of Traffic on Bridges: A Numerical Study . . . . . . . . . . . . . . . . . M. Sheibani, A.H. Hadjian-Shahri, and A.K. Ghorbani-Tanha

95

14

Covariance-Driven Stochastic Subspace Identification of an End-Supported Pontoon Bridge Under Varying Environmental Conditions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107 Knut Andreas Kvåle, Ole Øiseth, and Anders Rönnquist

vii

viii

Contents

15

Probabilistic Analysis of Human-Structure Interaction in the Vertical Direction for Pedestrian Bridges . . . . 117 Federica Tubino

16

Effects of Seismic Retrofit on the Dynamic Properties of a 4-Storey Parking Garage . . . . . . . . . . . . . . . . . . . . . . . . . . . 121 Ilaria Capraro and Carlos E. Ventura

17

Analytical and Experimental Study of Eddy Current Damper for Vibration Suppression in a Footbridge Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131 Wai Kei Ao and Paul Reynolds

18

Nonlinear Damping in Floor Vibrations Serviceability: Verification on a Laboratory Structure . . . . . . . . . . . . . . 139 Onur Avci

19

Addressing Parking Garage Vibrations for the Design of Research and Healthcare Facilities . . . . . . . . . . . . . . . . . 147 Brad Pridham, Nick Walters, Luke Nelson, and Brian Roeder

20

Modeling and Measurement of a Pedestrian’s Center-of-Mass Trajectory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159 Albert R. Ortiz, Bartlomiej Blachowski, Pawel Holobut, Jean M. Franco, Johannio Marulanda, and Peter Thomson

21

Evaluation of Mass-Spring-Damper Models for Dynamic Interaction Between Walking Humans and Civil Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169 Ahmed S. Mohammed and Aleksandar Pavic

22

Numerical Model for Human Induced Vibrations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179 Marcello Vanali, Marta Berardengo, and Stefano Manzoni

23

Dynamic Testing on the New Ticino Bridge of the A4 Highway. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187 Elena Mola, Franco Mola, Alfredo Cigada, and Giorgio Busca

24

Predicting Footbridge Vibrations Using a Probability-Based Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197 Lars Pedersen and Christian Frier

25

Flooring-Systems and Their Interaction with Usage of the Floor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205 Lars Pedersen, Christian Frier, and Lars Andersen

26

Benchmark Problem for Assessing Effects of Human-Structure Interaction in Footbridges . . . . . . . . . . . . . . . . . . . 213 S. Gómez, J. Marulanda, P. Thomson, J. J. García, D. Gómez, Albert R. Ortiz, S. J. Dyke, J. Caicedo, and S. Rietdyk

27

A Discrete-Time Feedforward-Feedback Compensator for Real-Time Hybrid Simulation. . . . . . . . . . . . . . . . . . . . . 223 Saeid Hayati and Wei Song

28

Sensing and Rating of Vehicle–Railroad Bridge Collision . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227 Shreya Vemuganti, Ali Ozdagli, Bideng Liu, Anela Bajric, Fernando Moreu, Matthew R. W. Brake, and Kevin Troyer

29

High-Frequency Impedance Measurements for Microsecond State Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235 Ryan A. Kettle, Jacob C. Dodson, and Steven R. Anton

30

Structural Stiffness Identification of Skewed Slab Bridges with Limited Information for Load Rating Purpose. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243 Abdollah Bagheri, Mohamad Alipour, Salman Usmani, Osman E. Ozbulut, and Devin K. Harris

31

Online Systems Parameters Identification for Structural Monitoring Using Algebraic Techniques . . . . . . . . . . . 251 L.G. Trujillo-Franco, G. Silva-Navarro, and F. Beltrán-Carbajal

32

Structural Vibration Control Using High Strength and Damping Capacity Shape Memory Alloys . . . . . . . . . . . 259 Soheil Saedi, Farzad S. Dizaji, Osman E. Ozbulut, and Haluk E. Karaca

33

Comparative Study on Modal Identification of a 10 Story RC Structure Using Free, Ambient and Forced Vibration Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267 Seyedsina Yousefianmoghadam, Andreas Stavridis, and Babak Moaveni

Contents

ix

34

Kronecker Product Formulation for System Identification of Discrete Convolution Filters . . . . . . . . . . . . . . . . . . . . 277 Lee Mazurek, Michael Harris, and Richard Christenson

35

Calibration-Free Footstep Frequency Estimation Using Structural Vibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287 Mostafa Mirshekari, Pei Zhang, and Hae Young Noh

36

Optimal Bridge Displacement Controlled by Train Speed on Real-Time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291 Piyush Garg, Ali Ozdagli, and Fernando Moreu

37

System Identification and Structural Modelling of Italian School Buildings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 301 Gerard O’Reilly, Ricardo Monteiro, Daniele Perrone, Igor Lanese, Matthew Fox, Alberto Pavese, and Andre Filiatrault

38

Investigation of Transmission of Pedestrian-Induced Vibration into a Vibration-Sensitive Experimental Facility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 305 Donald Nyawako, Paul Reynolds, Emma J. Hudson

39

An Ambient Vibration Test of an R/C Wall of an 18-Story Wood Building at the UBC Campus . . . . . . . . . . . . . . 315 Yavuz Kaya, Carlos E. Ventura, and Alireza Taale

40

The Day the Earth Shook: Controlling Construction-Induced Vibrations in Sensitive Occupancies . . . . . . . . . . 321 Michael J. Wesolowsky, Melissa W.Y. Wong, Todd A. Busch, and John C. Swallow

41

An Exploratory Study on Removing Environmental and Operational Effects Using a Regime-Switching Cointegration Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 329 Haichen Shi, Keith Worden, and Elizabeth J. Cross

42

Evaluation of Contemporary Guidelines for Floor Vibration Serviceability Assessment . . . . . . . . . . . . . . . . . . . . . . . 339 Zandy O. Muhammad, Paul Reynolds, and Emma J. Hudson

43

Excitation Energy Distribution of Measured Walking Forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 347 Atheer F. Hameed and Aleksandar Pavic

44

Identification of Human-Induced Loading Using a Joint Input-State Estimation Algorithm . . . . . . . . . . . . . . . . . . 353 Katrien Van Nimmen, Kristof Maes, Peter Van den Broeck, and Geert Lombaert

Chapter 1

Semi-Active Base Isolation of Civil Engineering Structures Based on Optimal Viscous Damping and Zero Dynamic Stiffness Felix Weber, Hans Distl, and Christian Braun

Abstract Spherical friction pendulums (FP) represent the common approach to isolate civil engineering structures against earthquake excitation. As these devices are passive and friction damping is nonlinear the optimal friction coefficient for minimum absolute acceleration of the building depends on the peak ground acceleration (PGA). Therefore, the common procedure is to optimize the friction coefficient for the PGA of the design basis earthquake (DBE) and to verify by simulations that the absolute structural acceleration for the maximum considered earthquake (MCE) is within a tolerable limit which is far from optimal. In order to overcome this drawback of passive FPs, a semi-active FP based on real-time controlled oil damper with the use of the collocated bearing displacement only is described in this paper. Four different semi-active control laws are presented that target to produce controlled dynamic stiffness depending on the actual bearing displacement amplitude in order to control the isolation period in real-time. The desired damping is formulated based on optimal viscous damping taking into account the passive lubricated friction of the spherical surface. The four control laws are compared in terms of absolute structural acceleration, bearing force, bearing displacement and residual bearing displacement. The results point out that the approach of zero dynamic stiffness at center position of the slider and nominal stiffness at design displacement of the FP improves the isolation of the structure within the entire PGA range significantly and at the same time minimize maximum bearing force, maximum bearing displacement and maximum residual bearing displacement. Keywords Control • Damping • Seismic • Semi-active • Negative stiffness

1.1 Introduction Spherical friction pendulums (FP) are widely used to significantly reduce the absolute structural acceleration due to ground excitation by their effective radius that shifts the fundamental time period of the isolated structure into the region of attenuation and their friction damping that augments the damping of the structure [1]. The inherent drawback of FPs is that friction damping is nonlinear whereby the optimal friction coefficient depends on the displacement amplitude of the FP and consequently peak ground acceleration (PGA) [2]. The common approach is therefore to optimize the friction coefficient for the PGA of the design basis earthquake (DBE) and, subsequently, to check if the absolute structural acceleration due to the maximum credible earthquake (MCE) is acceptable. In addition, it must be checked if the isolation of the structure at very small PGAs is acceptable from the comfort point of view since the constant friction coefficient being optimal for the PGA of the DBE may lead to clamping effects in the FP whereby the relative motion stops in the FP and consequently the structural absolute acceleration is equal to the ground acceleration. In order to overcome these drawbacks of FPs several types of adaptive FPs have been developed: FPs with several sliding surfaces with different friction coefficients and effective radii [3] and pendulums that are extended by an external active or semi-active actuator such as hydraulic cylinders and controllable dampers on the basis of oil dampers with controlled bypass valve or magnetorheological fluids [4–7]. Controllable dampers

F. Weber () Maurer Switzerland GmbH, Neptunstrasse 25, 8032 Zurich, Switzerland e-mail: [email protected] H. Distl Maurer Söhne Engineering GmbH & Co. KG, Frankfurter Ring 193, 80807 Munich, Germany e-mail: [email protected] C. Braun MAURER SE, Frankfurter Ring 193, 80807 Munich, Germany e-mail: [email protected] © The Society for Experimental Mechanics, Inc. 2017 J. Caicedo, S. Pakzad (eds.), Dynamics of Civil Structures, Volume 2, Conference Proceedings of the Society for Experimental Mechanics Series, DOI 10.1007/978-3-319-54777-0_1

1

2

F. Weber et al.

are seen to provide a promising solution as the resulting closed-loop is unconditionally stable and their power consumption is very low compared to hydraulic actuators. This paper describes a novel approach of a semi-active isolator with the following main features: • controlled dynamic stiffness depending on the actual displacement amplitude of the pendulum, • optimum viscous damping, and • collocated control based on one displacement sensor.

1.2 Systems Under Consideration 1.2.1 Friction Pendulum The common way to decouple the building/structure from the shaking ground is to support the building by FPs. The effective radius Reff D R  h of the FP is selected to shift the time period T of the non-isolated structure from the region of amplification, i.e. T is typically in the region 0.5–2.0 s, to the region of attenuation with associated isolation time period Tiso of typically 3–4 s. Subsequent to the design of the effective radius the friction coefficient  of the sliding surface is optimized for minimum absolute structural acceleration for given Tiso . As friction damping is nonlinear, the optimal value of  depends on the bearing displacement amplitude and consequently on PGA. As a result,  is commonly optimized for the PGA of the DBE. Finally, the structure with the designed FP is computed for the PGA due to the MCE to check if the absolute structural acceleration resulting from the MCE is acceptable and to know the displacement capacity of the FP that is required for the MCE.

1.2.2 Viscous Pendulum In addition to the passive FP an “ideal” pendulum without friction but with linear viscous damping is considered as benchmark for passive isolators. Its effective radius is equal to that of the FP to ensure the same isolation time period Tiso . Its viscous damping coefficient c is optimized for minimum absolute structural acceleration. Thanks to the linear behavior of viscous damping the optimization of c in independent of the bearing displacement amplitude and therefore independent of PGA.

1.2.3 Semi-Active Isolator The semi-active isolator consists of a passive FP and a semi-active damper that is installed between ground and top bearing plate of the pendulum (Fig. 1.1). The design of the effective radius will be explained in the section “CONTROL LAW” as it is related to the formulation of the control law. The sliding surface of the passive FP is lubricated to minimize the passive and therefore uncontrollable friction damping of the semi-active isolator and thereby to maximize the controllability of the total isolator force. The dissipative force of the semi-active damper is controlled by the electromagnetic bypass valve. The desired control force is computed by the real-time controller based on the measured bearing displacement which is identical to the relative motion between damper cylinder and damper piston. Based on the desired control force a force tracking module computes the valve command signal such that the actual force of the semi-active damper tracks closely its desired counterpart in real-time.

1.3 Modelling Due to the large isolation time period Tiso D 3.5 s of the building with isolator the building may be modelled as a single degree-of-freedom system [1]. The according equation of motion becomes ms uR s C cs .Pus  uP / C ks .us  u/ D ms uR g

(1.1)

1 Semi-Active Base Isolation of Civil Engineering Structures Based on Optimal Viscous Damping and Zero Dynamic Stiffness

3

us

..

ms

us

ks

ground

spherical hinge

cylinder

μ

piston

cs

m

semi-active damper valve

disp. sensor

R

controller

ground

..

ground

u

h

ug

ground

Fig. 1.1 Schematic of structure with semi-active isolator

where ms , cs , ks denote the modal mass, the viscous damping coefficient and the stiffness of the building, uR s , uP s and us denote the acceleration, velocity and displacement of the structure relative to the ground, uP and u are the velocity and displacement of the top bearing plate relative to the ground and uR g is the ground acceleration given by the accelerogram of the El Centro p North-South earthquake. The mass ms is determined by the typical vertical load of W D 6 MN on the isolator, cs D 2 s ks ms is computed based on the damping ratio  s D 1% and ks D 24.15 MN/m is selected such that the natural frequency of the building without isolator is 1 Hz representing a typical value for structures that require base isolation. The actual equation of motion of the top plate of the isolator with mass m and with actual force fsemiactive of the semi-active oil damper is m uR C fh C

W actual u D cs .Pus  uP / C ks .us  u/  fsemiactive  ms uR g Reff

(1.2)

where fh is the friction force of the curved sliding surface and W/Reff is the restoring stiffness due to the effective radius Reff D R  h of the pendulum. The force fh is modelled by the hysteretic damper modelling approach [8]  fh D

kh u W pre  sliding  W sign .Pu/ W sliding

(1.3)

where kh is the pre-sliding stiffness that is selected two orders of magnitude greater than W/Reff . In case of the passive pendulum without any friction but linear viscous damping fh in (1.2) is replaced by the term copt uP where copt denotes the optimal viscous damping coefficient of the isolator.

1.4 Control Law 1.4.1 General Formulation The desired active control force is formulated as follows    kcontrol u C copt  c uP W c  c  0 desired factive D W c  c < 0 kcontrol u

(1.4)

4

F. Weber et al.

in order to produce: 1. the controlled stiffness kcontrol that is controlled as function of the bearing displacement amplitude U to compensate for the passive stiffness of the curved surface given by W/Reff and thereby produce zero dynamic stiffness by kcontrol < 0 for maximum decoupling of the  from the ground, and  structure 2. the controlled damping force copt  c uP that dissipates the same amount of damping as resulting from optimal linear viscous damping. The desired optimal viscous damping coefficient copt is reduced by the viscous damping coefficient c that is energy equivalent to the friction damping of the lubricated curved surface [2] c 

4 W  !iso U

(1.5)

in order to dissipate the cycle energy of optimal viscous damping. Since c is inversely proportional to the displacement amplitude U of the isolator, i.e. c  U1 , c may become greater than copt at small U which necessitates the distinction of cases in (1.4). Notice that (1.5) represents an approximation because c according to Eq. (1.5) is derived based on the constant isolation radial frequency r !iso D

g Reff

(1.6)

but the actual frequency of the bearing displacement due to earthquake excitation is time-variant and therefore not detectable in real-time. However, the approximation (1.5) represents a good engineer’s solution as the actual frequency is in the vicinity of ! iso . The actual force of the semi-active oil damper can only produce the dissipative forces of the desired active desired , that is control force factive  actual fsemiactive

D

desired desired factive W uP factive 0 desired = 250 mm

fsemi-active / W (−)

0.06 0.04 0.02 0 −0.02 −0.04

positive stiffness due to U < 125 mm (plus kR-eff at U close to zero)

−0.06 −0.08 㻙300

0 −200 100 100 200 bearing displacement u (mm)

total isolator force f / W (−)

0.08

−0.10

control law #1 (CL #1) 0.10 zero dynamic stiffness during 1/2 cycle (with viscous damping superimposed) due to U >= 250 mm

0.06 0.04 0.02 0 −0.02 −0.04 positive stiffness due to U < 125 mm (kR-eff-nominal at U close to zero)

−0.06 −0.08 −0.10

300

−300

0 −200 100 100 200 bearing displacement u (mm)

300

Fig. 1.7 Force displacement trajectories of (a) semi-active control force and (b) total force of semi-active isolator due to control law #1 control law #2 (CL #2)

b

0.10 0.08 0.06

positive kR-eff during 1/2 cycle (with viscous damping superimposed) due to U >= 250 mm

fsemi-active / W (−)

0.04 0.02 0 −0.02 −0.04 −0.06

negative kR-eff during 1/2 cycle due to U close to zero

−0.08 −0.10

−300

−200

0 100 100 200 bearing displacement u (mm)

control law #2 (CL #2) 0.10 0.08

total isolator force f / W (−)

a

0.06

kR-eff-nominal during 1/2 cycle (with viscous damping superimposed) due to U >= 250 mm

0.04 0.02 0 −0.02 −0.04 −0.06

zero dynamic stiffness during 1/2 cycle at U close to zero

−0.08 300

−0.10

−300

−200

0 100 100 200 bearing displacement u (mm)

300

Fig. 1.8 Force displacement trajectories of (a) semi-active control force and (b) total force of semi-active isolator due to control law #2

compared to optimized friction pendulums and a hypothetical pendulum without friction but optimal viscous damping. This result is achieved without getting larger bearing displacements and forces and the re-centering requirement is also fulfilled semi-active base isolator. Acknowledgements The authors gratefully acknowledge the financial support of MAURER SE.

References 1. Tsai, C.S., Chiang, T.-C., Chen, B.-J.: Experimental evaluation of piecewise exact solution for predicting seismic responses of spherical sliding type isolated structures. Earthq. Eng. Struct. Dyn. 34, 1027–1046 (2005) 2. Weber, F., Boston, C.: Energy based optimization of viscous-friction dampers on cables. Smart Mater. Struct. 19, 045025 (11pp) (2010) 3. Fenz, D.M., Constantinou, M.C.: Spherical sliding isolation bearings with adaptive behavior: theory. Earthq. Eng. Struct. Dyn. 37, 163–183 (2008) 4. Feng, M.Q., Shinozuka, M., Fujii, S.: Friction-controllable sliding isolation system. J. Eng. Mech. (ASCE). 119(9), 1845–1864 (1993)

1 Semi-Active Base Isolation of Civil Engineering Structures Based on Optimal Viscous Damping and Zero Dynamic Stiffness

9

5. Kobori, T., Takahashi, M., Nasu, T., Niwa, N.: Seismic response controlled structure with active variable stiffness system. Earthq. Eng. Struct. Dyn. 22, 925–941 (1993) 6. Ramallo, J.C., Johnson, E.A., Spencer Jr., B.F.: ‘Smart’ base isolation systems. J. Eng. Mech. (ASCE). 128(10), 1088–1100 (2002) 7. Nagarajaiah, S., Sahasrabudhe, S.: Seismic response control of smart sliding isolated buildings using variable stiffness systems: an experimental and numerical study. Earthq. Eng. Struct. Dyn. 35(2), 177–197 (2006) 8. Ruderman, M.: Presliding hysteresis damping of LuGre and Maxwell-slip friction models. Mechatronics. 30, 225–230 (2015) 9. Preumont, A.: Vibration Control of Active Structures, Chapter 6. Kluwer Academic Publishers, Dordrecht (2002) 10. Weber, F., Ma´slanka, M.: Precise stiffness and damping emulation with MR dampers and its application to semi-active tuned mass dampers of Wolgograd Bridge. Smart Mater. Struct. 23, 015019 (2014)

Chapter 2

Long-Term Performance of Specialized Fluid Dampers Under Continuous Vibration on a Pedestrian Bridge Alan R. Klembczyk

Abstract In 2001, Taylor Devices Inc. developed special Viscous Dampers for use on the Millennium Bridge in London, England. These dampers were specified and designed to be used for mitigating the dynamic response of the bridge due to pedestrian traffic. Prior to the integration of the dampers, the bridge had experienced unacceptable movements, especially during periods when larger crowds of people were on the bridge. The result was that the bridge had to be closed until a solution was found. Much research was done and several papers were published about the nature of that problem and the ensuing solution. After successful component level testing and the installation of 37 Taylor Viscous Dampers, the bridge was re-opened to the public in February, 2002. Tests with approximately 2000 people demonstrated a much improved dynamic response. Since that time, the dampers have been subjected to almost constant dynamic input, some more than others. Due to the location of the bridge in central London, there has been nearly constant pedestrian traffic on the bridge each day and even throughout the night. However, because of the specialized nature of the damper design, no degradation in damper performance or in the dynamic response of the bridge itself has been experienced. This paper will outline the specifics in quantifying the continued damper performance through an intermediate inspection after 7 years, followed by a successful comprehensive inspection after 11 years. This included the removal, dynamic testing, and re-installation of three selected dampers. Keywords Millennium Bridge • Bridge damper test results • Fluid viscous dampers • Continuous vibration • Vibration damper

2.1 Introduction The unique design and the resulting unacceptable response of the Millennium Bridge in central London (see Fig. 2.1) have been well publicized and documented. The specifics of this dynamic response and the resulting solution will not be reiterated within the context of this paper. However, in order to provide a necessary background, a short summary is presented here. In June 2000, the bridge was first opened to the public. Shortly thereafter, with substantial pedestrian traffic present, the bridge began to sway in a lateral motion to the discomfort of many of the pedestrians. The bridge was subsequently shut down and significant studies were performed to provide solutions to stop the excessive swaying. Since the response frequency was near the frequency of human footfalls during walking, it was determined that stiffening of the structure was not a practical solution. The unique design and its aesthetic appearance would have been sacrificed if structural modifications were made to keep the various modal frequencies away from walking frequencies. A more acceptable solution was determined to substantially increase the damping level of the bridge over all input conditions in order to prevent pedestrian traffic from exciting the bridge. The required amount of added damping was determined to be nearly 20% critical, a value that is effectively unachievable with typical solutions, such as tuned mass dampers, frictional elements, or structural modifications. Many challenges became immediately apparent when proposing a damping solution for this unique structure. One of the most significant was the fact that the owner of the bridge required a permanent and maintenance-free solution that would last throughout the life of the bridge; this being in excess of 50 years. Since the expected pedestrian traffic was such that the dampers would cycle nearly continuously at 1.3 Hz, it was necessary to specify a cycle life of 2  109 cycles minimum. Due to this stringent requirement, Taylor Devices proposed the use of specialized Fluid Dampers that employed the use of flexing metal bellows seals, rather than traditional sliding seals that are elastomeric in nature and therefore subject to wear and degradation over long-term environmental and cyclic conditions.

A.R. Klembczyk () Taylor Devices, Inc., 90 Taylor Drive, North Tonawanda, NY, 14120-0748, USA e-mail: [email protected] © The Society for Experimental Mechanics, Inc. 2017 J. Caicedo, S. Pakzad (eds.), Dynamics of Civil Structures, Volume 2, Conference Proceedings of the Society for Experimental Mechanics Series, DOI 10.1007/978-3-319-54777-0_2

11

12

A.R. Klembczyk

Fig. 2.1 The Millennium Bridge

2.2 Specialized Damper Design [1] Taylor Devices’ Fluid Dampers with metal bellows seals had been previously used exclusively by NASA and other U.S. Government agencies for space based optical systems. These previous applications had similar requirements for long life and high resolution at low amplitudes, but required relatively low damper forces from small, lightweight design envelopes. Figure 2.2 is a photograph of a pair of typical dampers of this design, used in space on more than 70 satellites to protect delicate solar array panels. This figure also shows the metal bellows seals; one in the compressed position and one in the extended position. This type of seal does not slide, but rather flexes without hysteresis as the damper moves. This patented design is known as a Frictionless Hermetic Damper. A cutaway of a typical damper of this type is shown in Fig. 2.3. Two metal bellows seals are used to seal fluid in each damper, one at each end of the damping chamber. As the damper moves, the two metal bellows alternately extend and retract, by flexure of the individual bellows segments. Since the seal element elastically flexes rather than slides, seal hysteresis is nearly zero. The volume displaced by the compressing bellows passes through the crossover ports to the extending bellows at the opposite end of the damper. While this is occurring, damping forces are being produced by orifices in the damping head, and the pressures generated are kept isolated from the metal bellows by high restriction hydrodynamic labyrinth bushings. Because hydrodynamic bushings are used, no sliding contact with the piston rod occurs, assuring near-frictionless performance. Adapting this basic design for use on the Millennium Bridge largely involved simply scaling the small satellite Dampers to the required size range. All parts, including the metal bellows seals, were designed with low stress levels to provide an endurance life in excess of 2  109 cycles. The metal bellows and other moving parts were constructed from stainless steel for corrosion resistance. To assure a high resolution output, it was required that all damper attachment clevises be fabricated with fitted spherical bearings and fitted mounting pins, such that zero net end play existed in the attachment brackets. A total of 37 dampers of this design were manufactured, component-level tested, and installed on the bridge in late 2001. There are three basic types of dampers. These are referred to as the Pier Dampers, the Deck Dampers, and the Vertical Dampers and are described below: Damper Nomenclature: Pier Damper Quantity on the Bridge: 16 Description: Two Pier Dampers are located on each side of each of two piers on both the east and west side of the bridge, for a total of eight dampers per pier. Damping coefficient values for the eight dampers connected directly to the center span of the bridge are significantly higher than the other Pier Dampers. Dampers have varying over-all lengths due to the

2 Long-Term Performance of Specialized Fluid Dampers Under Continuous Vibration on a Pedestrian Bridge

13

Fig. 2.2 Space satellite dampers

Fig. 2.3 Cutaway of frictionless hermetic damper

location of the attachment points, the longest being 8.3 m long. These dampers are quite apparent to pedestrians when crossing the bridge as illustrated in Figs. 2.4 and 2.5 below. Damper Nomenclature: Deck Damper Quantity on the Bridge: 17 Description: The Deck Dampers are located under various deck sections. A very limited number can be seen from under the north end of the bridge. Most deck dampers are not visible since they are situated directly under the deck panels. Lateral motions of the bridge are transmitted to the dampers through pairs of relatively long V-shaped chevron braces as shown in Figs. 2.6 and 2.7 below. Damper Nomenclature: Vertical Damper Quantity on the Bridge: 4 Description: Vertical Dampers are located in two pairs under the south end of the bridge with damper ends connected between a structural arm and the ground. As illustrated below in Figs. 2.8 and 2.9, the dampers are directly accessible to pedestrian traffic. Nearly continuous damped motion is felt and observed with even low to moderate pedestrian traffic on the bridge overhead.

14

Fig. 2.4 4 of 16 pier dampers

Fig. 2.5 Moving end of pier damper over the River Thames

Fig. 2.6 Deck damper shown with chevron connection

A.R. Klembczyk

2 Long-Term Performance of Specialized Fluid Dampers Under Continuous Vibration on a Pedestrian Bridge

Fig. 2.7 Deck panels removed deck damper showing

Fig. 2.8 Inspection of vertical damper pair

Fig. 2.9 Vertical damper pair with pedestrian access

15

16

A.R. Klembczyk

2.3 Intermediate Inspection After 7 Years in Service A visual inspection of each damper was performed looking for corrosion, damage to the unit from use or the surrounding environment, and for fluid leakage. The units were all found to be in 100% working condition with minimal signs of physical damage or deterioration, as well as no signs of fluid leakage. There were only minor signs of corrosion and some external contamination noted. The units had been subjected to nearly constant cycling for a period of use of over 7 years at the time of this inspection. The total estimated cycles after 7 years was as many as 2.0  108 . The owner required no formal testing of installed dampers at this time.

2.4 Principal Inspection and Testing After 11 Years in Service The Principal Inspection after 11 years of service included two phases. The first was a visual inspection of all Pier Dampers and all four Vertical Dampers. All dampers appeared to be in 100% working order. A sample of five of the seventeen deck dampers were inspected per the owner’s request to minimize deck panel removal costs. Similar to the case for the Intermediate Inspection 4 years earlier, there were only minor signs of corrosion and some external contamination noted. This minor corrosion and contamination appears to have been caused by caustic chemicals from the exhaust plumes from boats and ships navigating under the bridge. Dampers located under the deck of the bridge near the shore or over land exhibited nearly new appearance. Two of the five Deck Dampers and one of the four Vertical Dampers were temporarily removed for testing purposes as outlined below. The second phase of the Principal Inspection consisted of performing dynamic tests on the three dampers that were removed. These three dampers were shipped to the Taylor Devices facility in North Tonawanda, New York so that they could be tested to the original Acceptance Test Procedure and compared to the original acceptance tests from 2001. This was done to determine if any of the performance outputs had deteriorated in any way. This Acceptance Test Procedure consisted of two types of tests. The first type consisted of subjecting the dampers to a series of sinusoidal input tests throughout the specified velocity range. These tests are referred to as the “Force vs. Velocity” tests. The second type of test was performed at approximately 0.50 mm amplitude. These tests are referred to as the “Low Amplitude” tests. The Low Amplitude test demonstrates the ability of each Damper to produce substantial damping force for very small vibrations, and demonstrate that there has been no loss of fluid. If any loss of fluid had occurred, the damper would demonstrate an inability to produce any substantial force for these small displacements. Figures 2.10, 2.11, and 2.12 show the results of the Force versus Velocity tests for each Damper, measuring the output force at several velocity inputs. These plots also show the data points recorded through the same testing methods 11 years prior. The graphical data illustrates the fact that there is virtually no difference in output characteristics when comparing the results from 2001 to the results from 2012. Figures 2.13, 2.14 and 2.15 demonstrate the results of the Low Amplitude Tests for each of the three dampers that were tested. Note that in each case, the hysteresis loops (force vs. displacement) show no signs of free-play, loss of fluid, excessive friction, wear or degradation of any sort. It should be noted that the dampers were tested with their spherical bearings in place and their end attachment brackets still connected. Therefore, no degradation to these components has occurred and the bearings have maintained their tight fit requirement that is necessary to produce damping for very low displacements. Subsequent to the successful testing of these three dampers, they were sent back to London and reinstalled on the bridge in January 2013.

2.5 Conclusions The results of the 7-year Intermediate Inspection, the 11-year Principal Inspection, and dynamic testing show that the Millennium Bridge dampers have experienced no physical or functional deterioration. The dampers displayed no measurable change in output, as well as no signs of leakage after 11 years of continuous service and nearly constant cycling.

2 Long-Term Performance of Specialized Fluid Dampers Under Continuous Vibration on a Pedestrian Bridge

Fig. 2.10 Force vs. Velocity test results of deck damper 2001 & 2012

Fig. 2.11 Force vs. Velocity test results of deck damper 2001 & 2012

17

18

A.R. Klembczyk

Fig. 2.12 Force vs. Velocity test results of vertical damper 2001 & 2012

5.0

Force (kN)

2.5

0.0

-2.5

-5.0 -1.00

-0.75

-0.50

Fig. 2.13 Low amplitude test results of deck damper 2012

-0.25

0.00 0.25 Stroke (mm)

0.50

0.75

1.00

2 Long-Term Performance of Specialized Fluid Dampers Under Continuous Vibration on a Pedestrian Bridge

19

10.0 7.5 5.0

Force (kN)

2.5 0.0 -2.5 -5.0 -7.5 -10.0 -1.25

-1.00

-0.75

-0.50

-0.25 0.00 0.25 Stroke (mm)

0.50

0.75

1.00

1.25

Fig. 2.14 Low amplitude test results of deck damper 2012

Force (kN)

2.5

0.0

-2.5 -1.00

-0.75

-0.50

Fig. 2.15 Low amplitude test results of vertical damper 2012

-0.25

0.00 0.25 Stroke (mm)

0.50

0.75

1.00

20

A.R. Klembczyk

The dampers were originally designed and built for this nearly constant cycling over a period of more than 50 years, projected to total approximately 2  109 (2 billion) cycles. Due to the fact that the results of the intermediate and principal inspections and testing show no signs of degradation, it is anticipated that the dampers will be able to meet this expected life time as anticipated. Acknowledgements The author would like to thank Craig Winters, Robert Schneider and Sean Frye for their contributions to this paper.

Reference 1. Taylor, D.: Damper retrofit of the London Millennium Footbridge—a case study in biodynamic design. In: Proceedings of the 73rd Shock and Vibration Symposium, 2002

Chapter 3

Analysis of Variation Rate of Displacement to Temperature of Service Stage Cable-Stayed Bridge Using Temperatures and Displacement Data Hyun-Joong Kim

Abstract Because stage cable-bridge have long spans and large members, their movements and geometrical changes by temperatures tend to be bigger than those of small or medium-size bridge. Therefore, it is important for maintenance engineers to monitor and assess the effect to temperature on the cable-supported bridges. To evaluate how much the superstructure expands or contracts when subjected to changes in temperature is the first step for the maintenance. Thermal movements of a cable-stayed bridge in service are evaluated by using long-term temperatures and displacements data. Keywords Cable-stayed bridge • Thermal movements • Temperature effect • Correlation analysis

3.1 Introduction Since cable-supported bridges are long and sizable, their movement and transportation amount are quite considerable compared to those of general bridges. Thus, in order for the maintenance of cable-supported bridges, it is important to analyze and evaluate the influence of temperature on the existing bridges; in terms of such temperature evaluation, assessing elastic superstructures of bridges depending on the change in temperature is known to be the most basic analyzing techniques. Recently, there are a variety of ongoing research projects to evaluate the structures more precisely as the existing period of cable-supported bridges getting extended and the amount of their accumulated data is increasing [2–4]. In our research, in order to study the influence of temperature on cable-supported bridges, we employed the previous methods such as using an expansion displacement meter or single measuring method with data gathered from Global Navigation Satellite System (GNSS). Then, based on the cases that evaluated temperature expansion behaviors using different datasets to enhance their statistical reliability, we measured the expansion per unit temperature using the temperature data and the GNSS displacement data of existing cable bridges and confirmed the structural integrity of the bridges through the comparison between our results and theoretical values.

3.2 The Subject Bridge and Data Collection Method 3.2.1 Subject Bridge The bridge used for this research is Cheong-Poong Dae-Kyo (Bridge) located in Jechun, Choong-Cheong-Book-Do, which is a 5-consecutive-span complex cable-stayed bridge with the full length of 472 m (center: 327 m, side span: 57.5 m, abutment: 30 m) and width of 13 m and has RC cross sections for its side spans and composite cross sections consisting of precast concrete floor slap and I-shaped structural steel cross sections for its main spans. Its tower has a H-shaped concrete cross section with the height of 103 m and was completed in May 2012; it is currently under the management of Structural Health Monitoring System (SHMS).

H.-J. Kim () EJ Tech Co., Ltd., Seokjeong B/D, 10, migeum-ro 33 beon-gil, Bundang-gu, Seongnam-si, Gyeonggi-do, Republic of Korea e-mail: [email protected] © The Society for Experimental Mechanics, Inc. 2017 J. Caicedo, S. Pakzad (eds.), Dynamics of Civil Structures, Volume 2, Conference Proceedings of the Society for Experimental Mechanics Series, DOI 10.1007/978-3-319-54777-0_3

21

22

H.-J. Kim

Fig. 3.1 Installation place of GNSS and thermometer

3.2.2 Data Collection Method We collected our data using the sensors utilized to measure the expansion of Cheong-Poong Dae-Kyo in the 2 receivers in the central girder and the rest 4 GNSSes except for 2 reference stations among the 8 previously installed GNSS receivers as shown in Fig. 3.1 and also the member thermometer installed at the ¼ location of the stiffner. The duration of our data collection was a year, from January 1st, 2014 to December 31st, 2014, and we performed a correlation analysis of GNSS displacements along the direction of the bridge using the 10-min average data of GNSS. The GNSS data consists of three components, dN, dE, and dH, and we defined directions as “perpendicular” and “corresponding” to the bridge line by turning the azimuth 22ı counterclockwise when our Y-axis is due north (N) and X-axis due east (E).

3.2.3 The Analysis of Expansion per Unit Temperature The amount of expansion due to the temperature change of a structure is calculated as shown in the formula (3.1). Here, L implies the temperature change, L the expansion length, and ’ the coefficient of linear expansion of a material. In the formula (3.1), ’ and L become the values for the boundary conditions depending on the used material and the member length, respectively. We can see that this is a linear relationship between the expansion amount and temperature and define it as the formula (3.2) shown below. ST is defined as the expansion per unit temperature (mm/ı C) [1] LT D ’  T  L

(3.1)

LT =T D ’  L D ST

(3.2)

The calculation of expansion per unit temperature using measured data is to estimate the linear regression (3.3) from the measured values, ˇ 0 , ˇ 1 of two variables using the least square method (3.4) and its formula is presented below. y D ˇ0 C ˇ1 x   P   xi  x yi  y Sxy b ˇ1 D D  2 P  Sxx xi  x

(3.3)

(3.4)

3 Analysis of Variation Rate of Displacement to Temperature of Service Stage. . .

(a)

(b)

GPS(p2-L)

80

20 0 −20 −40

40 30 20 10 0

−10

−60

−20 −10

−5

0

(c)

5

10 15 20 Temperature (°C)

25

30

35

−30 −15

40

(d)

GPS(p2-L)

80

−10

−5

0

5

10 15 20 Temperature (°C)

Displacement (mm)

20 0 −20 −40

30

35

40

y = −1.831x + 41.07 R2 = 0.848

50

40

25

GPS(PY1-L)

70 60

y = −2.323x + 35.64 R2 = 0.923

60 Displacement (mm)

y = −1.831x + 41.07 R2 = 0.848

50 Displacement (mm)

Displacement (mm)

60

y = −2.323x + 35.64 R2 = 0.923

40

40 30 20 10 0

−10

−60 −80 −15

GPS(PY1-L)

70

60

−80 −15

23

−20 −10

−5

0

5

10 15 20 Temperature (°C)

25

30

35

40

−30 −15

−10

−5

0

5

10 15 20 Temperature (°C)

25

30

35

40

Fig. 3.2 Relationship between air temperatures and displacement. (a) Relationship between air temperatures and P2. (b) Relationship between air temperatures and PY1. (c) Relationship between air temperatures and PY2. (d) Relationship between air temperatures and A2

Sxy D

X

xi yi 

Sxx D

X

X  X  yi =n xi

(3.5)

X 2 xi =n

(3.6)

xi2 

3.3 The Analysis Through the analysis of measured data, we confirmed a high correlation between two variables as shown in Fig. 3.2 by setting the effective temperature considering expansion directions of bridges and the expansion measured at each location of the bridge as two variables and displayed their distribution [5–8]. We found our coefficient of linear expansion as 0.00001 after calculating the slope of the linear regression equation as displayed in Fig. 3.3. We estimated the theoretical expansion amount per unit temperature as shown in Table 3.1 after applying the coefficient of linear expansion analyzed through our measurement, put it into the formula (3.4), and compared the expansion amounts per unit temperature of temperature and displacement data as presented in Table 3.2.

24

H.-J. Kim

Unut themal movement (mm/°C)

3

y = 0.010x - 0.108 R2 = 0.999

2

1

0 −300

−200

−100

0

200

100

300

−1 −2 −3 Distance (m)

Fig. 3.3 Relationship between expansion length and ST LT T LT T LT T LT T

Table 3.1 Analytical comparison of variation rate of displacement to temperature

P2 PY1 PY2 A2

Table 3.2 Comparison of variation rate of displacement to temperature

Location L (m) ST (mm/ı C) Measurement Theory

D ’  LP2 D 0:0001  221 D 2:21 mm=ı C D ’  LPY1 D 0:0001  163:5 D 1:635 mm=ı C D ’  LPY2 D 0:0001  163:5 D 1:635 mm=ı C D ’  LA2 D 0:0001  221 D 2:21 mm=ı C P2 221

PY1 163.5

PY2 163.5

A2 221

2.323 2.21

1.831 1.635

1.593 1.635

2.13 2.21

3.4 Conclusion After analyzing correlation (from January 1st, 2014 to December 31st, 2014) through the temperature change and GNSS bridge displacement data, we found a high linear correlation between two variables and confirmed that the coefficient of linear expansion of Cheong-Poong bridge was 1.00E-05/ı C. In addition, as we compared the theoretical expansion amount per unit temperature with the measured expansion amount, we found a similar result. After considering the previous research results that a sound structure system showed a linear correlation with strong elastic behaviors depending on the change in temperature and comparing the expansion amount per unit temperature that can evaluate its soundness, we assessed that the structure system and long-term temperature expansion behaviors of Cheong-Poong bridge were quite sound. In order for a precise evaluation of soundness in the future, we expect that a multidimensional analysis of different temperature datasets would be necessary. Acknowledgments This work is a part of a research project supported by Ministry of Science, ICT and Future Planning, Institute for Information & communications Technology Promotion (IITP) through Core Research Project No. R-20160216-002659. The authors wish to express their gratitude for the financial support.

References 1. American Association of State Highway and Transportation Officials (AASHTO): AASHTO LRFD bridge design specifications. Washington (2012) 2. Bae, I.H.: Measurement analysis and management of special bridges. In: Annual Seminar on the Operation and Maintenance of Cable-Supported Bridges in Korea, The New Airport Highway Corporation (in Korean) (2011)

3 Analysis of Variation Rate of Displacement to Temperature of Service Stage. . .

25

3. Bae, I.H., Choi, B.K., Na, W.C.: Analysis of long-term behaviors and emergency measures of long span cable-supported bridge using the structural health monitoring system. In: Proceedings of Korean Society of Civil Engineers Convention, pp. 876–879 (2013) 4. Naoaki, S.: Long term monitoring in Akashi Kaikyo Bridge. Bridg. Found. (in Japanese). 37(6), 21–25 (2003) 5. Ni, Y.Q., Hua, X.G., Wong, K.Y., Ko, J.M.: Assessment of bridge expansion joints using long-term displacement and temperature measurement. J. Perform. Constr. Facil. 21(2), 143–151 (2007) 6. Xu, Y.L., Chen, B., Ng, C.L., Wong, K.Y., Chan, W.Y.: Monitoring temperature effect on a long suspension bridge. Struct. Control. Health Monit. 17(6), 632–653 (2010) 7. Zhou, Y., Sun, L.: Temperature effects on performance of long-span cable-stayed bridges. In: The 6th World Conference on Structural Control and Monitoring, Barcelona, Spain, pp. 432–444 (2014) 8. Park, J.C.: Evaluation of thermal movements of a cable-stayed bridge using temperatures and displacements data. J. Kor. Soc. Civil Eng. 35(4), 779–789 (2015)

Chapter 4

Triple Friction Pendulum: Does It Improve the Isolation Performance? Felix Weber, Peter Huber, Hans Distl, and Christian Braun

Abstract The working philosophy of the triple friction pendulum (FP) is to generate low friction in the region of 1.5–2% combined with high stiffness due to the small effective radii of the articulated slider assembly at low peak ground accelerations (PGAs), to produce medium friction around 3–5% and medium stiffness by simultaneous sliding on surfaces 1 and 4 at PGAs corresponding to the design basis earthquake (DBE), to generate increasing friction with further increasing PGAs up to the maximum credible earthquake (MCE) by the high friction of surface 4 in the region of 10% and, eventually, to produce considerably increased stiffness at PGAs beyond of MCE in order to reduce the maximum required displacement capacity of the triple FP. This study first investigates if this design philosophy results in enhanced isolation of the primary structure compared to the conventional FP. For this, the triple FP according to the above mentioned design concept is numerically tested for several earthquakes that are scaled to various PGAs in order to operate the triple FP within all its sliding regimes with associated isolation efficiencies. These results are compared to those of the conventional non-adaptive double FP with equal friction coefficients on its sliding surfaces and same effective radii as those of concave plates 1 and 4 of the triple FP to ensure equal isolation time periods. The first study demonstrates that the conventional FP outperforms the triple FP for most of the PGAs except for very small PGAs below 1–2 m/s2 depending on the earthquake. This finding is explained by the facts that the small effective radii of the articulated slider assembly reduce the isolation time period and therefore the isolation of the structure and splitting the friction into medium and high values on surfaces 1 and 4 cannot improve the isolation performance since the relative motion amplitudes on surfaces 1 and 4 are reverse whereby the energy dissipation is not enhanced compared to the conventional double FP with equal friction coefficients. The second study shows a way how the triple FP with four friction coefficients, four effective radii and four displacement capacities can be optimized for maximum isolation of the primary structure. This study points out that the optimized triple FP converges to the optimized double FP which explains the similar isolation performances. Thus, the triple FP does not improve the isolation of the structure compared to the conventional friction pendulum. Keywords Curved surface slider • Earthquake • Friction • Isolation • Pendulum • Seismic

4.1 Introduction Conventional spherical friction pendulums (FP) such as single and double FPs are often used isolator. The triple FP with the articulated slider assembly in between of the two main concave sliding surfaces has become famous because it generates displacement amplitude dependent stiffness and friction behaviors [1, 2]. According to [1, 2] the triple FP is intended to produce low friction and high stiffness at small bearing motion amplitudes and peak ground accelerations (PGA), respectively, to exert increasing friction at significantly reduced stiffness at medium bearing displacement amplitudes and PGAs, respectively, due to the design basis earthquake (DBE), to generate further increasing friction at further lowered stiffness at large bearing displacement amplitudes and PGAs, respectively, due to earthquakes between DBE and the

F. Weber () Maurer Switzerland GmbH, Neptunstrasse 25, 8032 Zurich, Switzerland e-mail: [email protected] P. Huber • C. Braun MAURER SE, Frankfurter Ring 193, 80807 Munich, Germany e-mail: [email protected]; [email protected] H. Distl Maurer Söhne Engineering GmbH & Co. KG, Frankfurter Ring 193, 80807 Munich, Germany e-mail: [email protected] © The Society for Experimental Mechanics, Inc. 2017 J. Caicedo, S. Pakzad (eds.), Dynamics of Civil Structures, Volume 2, Conference Proceedings of the Society for Experimental Mechanics Series, DOI 10.1007/978-3-319-54777-0_4

27

28

F. Weber et al.

a

R4, m4 R3, m3

b

d3

4 3 h4 h1

h3 h2

d4

2

2

u3

h2

u2

2 d1

1

(ground) R2, m2

1

h1

u1

d2 d1

u2=utot u1 üg

(ground)

üg R1, m1

R2, m2

u4=utot

R1, m1

d2

Fig. 4.1 Sketches of triple FP (a) with articulated slider assembly and double FP (b) without articulated slider

maximum credible earthquake (MCE), and to exhibit stiffening behavior for earthquakes beyond of MCE to limit the required maximum displacement capacity. This paper investigates if the above mentioned behavior improves the isolation of the structure.

4.2 Friction Pendulums Under Consideration 4.2.1 Triple Friction Pendulum The triple FP consists of the articulated slider assembly with sliding surfaces 2 and 3, the articulated slider in between and the main sliding surfaces 1 and 4 (Fig. 4.1a). The common design is that the effective radii 2 and 3 (Reff  2 , Reff  3 ) are equal and approx. 8 times smaller than the equal effective radii 1 and 4 (Reff  1 , Reff  4 ) that determine the isolation time period of the bearing s Tiso D 2 

Reff 1 C Reff 4 g

(4.1)

The friction coefficients 2 and 3 are usually selected to be small in the range of 1.5–2% while 1 is designed to produce medium friction around 3–5% and 4 high friction in the region of 8–11%. The restrainers on sliding surfaces 2 and 3 are required to initiate sliding on surfaces 1 and 4 and consequently always included in the triple FP. In contrast, the restrainers on surfaces 1 and 4 are not always included but are needed when the intended stiffening behavior due to sliding regime V for earthquakes beyond of MCE is desired and when the structural engineer specifies that the bearing must include end stoppers. For the first study of this paper the triple FP is designed according to the published design [1, 2] which represents a mock-up triple FP: Reff  1 D Reff  4 D 0.435 m, Reff  2 D Reff  3 D 0.053 m, 1 D 3.1%, 2 D 3 D 1.75% and 4 D 11.4% where the friction values represent average values identified from testing [2]. The displacement capacities are d1 D d4 D0.064 m and d2 D d3 D 0.019 m that yield the total displacement capacity dtot D 0.166 m. For the second study of this paper the parameters of the triple FP are optimized for minimum absolute structural acceleration which is described in the penultimate section.

4.2.2 Double Friction Pendulum The conventional, i.e. non-adaptive, double FP is composed of two bearing plates with equal effective radii Reff  1 D Reff  2 and friction coefficients 1 and 2 that must be equal because the relative motions on surfaces 1 (u1 ) and 2 (u2 ) are coupled due to the non-articulated slider (Fig. 4.1b). For the first study the isolation time period given by the sum (Reff  1 C Reff  2 ) is equal to that of the triple FP (1), the equal friction coefficients 1 D 2 are tuned by trial and error and dtot is the same as for the mock-up triple FP. For the second study of this paper a double FP with articulated slider is assumed which allows optimizing the double FP by different 1 and 2 .

4 Triple Friction Pendulum: Does It Improve the Isolation Performance?

29

4.3 Assessment Criterion The isolation performances of the triple FP and the non-adaptive double FP as benchmark are assessed in terms of the extreme value of the absolute acceleration of the structure   (4.2) max jRus C uR g j where uR s denotes the acceleration of the structure relative to the shaking ground and uR g is the ground acceleration given by the El Centro North-South (NS), the Kobe, the Loma Prieta and the Northridge earthquakes. In order to operate the triple FP within all its sliding regimes with associated different isolation performances the above mentioned accelerograms are scaled to the following PGAs PGA D Œ0:5 W 0:5 W PGAmax 

(4.3)

where PGAmax describes this PGA value at which the full displacement capacity of the triple FP is used.

4.4 Modelling 4.4.1 Approach In order to compute the maximum acceleration response of the structure with triple and double FPs the coupled nonlinear (friction damping) equations of motion for the lumped mass of the building and the masses of all bearing plates are ® formulated. The resulting stiff system of nonlinear differential equations is solved in the time domain in Matlab using the solver ode15s(stiff/NDF) with maximum relative tolerance 1e-3 and variable step size with upper bound of 1e-5 s. The excitation force is given by the PGA-scaled accelerograms of the aforementioned earthquakes. Subsequent to the dynamic simulations the assessment criterion (4.2) is determined from the time histories of the absolute structural acceleration.

4.4.2 Coupled Nonlinear Equations of Motion The building with isolator is modelled as a single degree-of-freedom system ms uR s C cs .Pus  uP 4 / C ks .us  u4 / D ms uR g

(4.4)

where ms , cs , ks denote the modal mass, the viscous damping coefficient and the stiffness of the building, uR s , uP s and us denote the acceleration, velocity and displacement of the structure relative to the ground while uP 4 and u4 are the velocity and displacement of the top bearing plate of the triple FP relative to the ground. cs in (4.4) is computed based on the damping ratio  s D 1%. For study with the mock-up triple FP ks is selected so that the isolation time period (1), which is 1.87 s, is two times higher than the natural period of the non-isolated structure; for the second study the isolation time period is set to the typical value of 3.5 s in order to shift the structure out of the frequency range of most earthquakes and the natural period of the non-isolated structure is set to the typical value of 1.2 s. The equation of motion of concave plate 4 of the triple FP with mass m4 becomes m4 uR 4 C fh4 C

W .u4  u3 / C fr4 D cs .Pus  uP 4 / C ks .us  u4 /  m4 uR g Reff 4

(4.5)

where fh  4 is the friction force of the curved sliding surface, W/Reff  4 is the restoring stiffness due to the vertical load W D 9.81 ms on the bearing and the effective radius Reff  4 D R4  h4 ; uP 3 and u3 are the velocity and displacement of bearing plate 3 of the triple FP relative to the ground. The force fh  4 is modelled by the hysteretic damper modelling approach [3]  fh4 D

kh4 .u4  u3 / W pre  sliding sliding 4 W sign .Pu4  uP 3 / W

(4.6)

30

F. Weber et al.

where kh  4 is the pre-sliding stiffness that is selected two orders of magnitude greater than W/Reff  4 . The force of the restrainer of concave plate 4 is assumed as linear stiffness force if it is triggered  fr4 D

kr4 .ju4  u3 j  d4 / sign .u4  u3 / W ju4  u3 j  d4 0 W ju4  u3 j < d4

(4.7)

where kh  4 denotes the restrainer stiffness which is also assumed to be two orders of magnitude greater than W/Reff  4 . The equations of motion of concave plate 3 (i D 3) and slider mass 2 (i D 2) have the same form mi uR i C fhi C

W Reff i

.ui  ui1 / C fri D fh.iC1/ C

W Reff .iC1/

.uiC1  ui / C fr.iC1/  mi uR g

(4.8)

where the friction and restrainer forces fh  i and fr  i are formulated analogically with (4.6) and (4.7). The equation of motion of concave plate 1 is given by Eq. (4.8) with i D 1 and ui1 D uP i1 D 0. The equation of motion of concave plate 2 of the double FP is given analogically with (4.5) and the equation of motion of the slider (index 1) of the double FP is given by (4.8) with i D 1 and ui1 D uP i1 D 0.

4.5 Isolation Performance of Mock-Up Triple Friction Pendulum The isolation results of the triple FP that is designed according to the literature [1, 2] and of the double FP with same isolation time period and equal friction coefficients 1 D 2 that are tuned by trial and error are depicted in Figs. 4.2 and 4.3 for the four chosen accelerograms. The following main observation can be made: • The triple FP generates a worse isolation of the structure for most of the PGAs (PGA2.2 m/s2 for El Centro NS, PGA0.59 m/s2 for Kobe, PGA0.85 m/s2 for Loma Prieta, PGA1.12 m/s2 for Northridge) than the double FP whose equal friction coefficients are tuned by trial and error (suboptimal tuning). • The isolation due to the triple FP deteriorates dramatically when sliding regime V is triggered due to the stiffening behavior at reduced friction of sliding regime V [4]. • The fact that the restrainers of the double FP with same total displacement capacity are triggered at larger PGAs than in case of the triple FP demonstrates that the double FP does not only generate better isolation of the structure but also that the total bearing displacement is smaller.

El Centro NS triple FP double FP, μ1=μ2=6.0% double FP, μ1=μ2=6.5% without isolator (max=19.1m/s2)

restrainers triggered (triple and double FPs)

triple FP, μ1=4%, μ4=13% triple FP, μ1=5%, μ4=15%

5 4

4.5

sliding regime V triggered sliding regime V not triggered

3

2.2

2

1

2

3.5

3

5

4 2

PGA (m/s )

6

7

restrainers triggered (triple and double FPs)

sliding regime V triggered

3 sliding regime V not triggered

2.5 2 1.5

0.5

7.8

restrainers triggered 0

4

1

1 0

triple FP double FP, μ1=μ2=5.5% double FP, μ1=μ2=6.5% without isolator (max=12.22m/s2)

5

3.82

6

Kobe 5.5

0.59

7

max( | üs+üg | ) (m/s2)

b

8

max( | üs+üg | ) (m/s2)

a

8

0

0

0.5

1

1.5

2.5

2

3

3.5

4

2

PGA (m/s )

Fig. 4.2 Isolation performances of triple FP according to literature and conventional double FP tuned by trial and error for El Centro NS (a) and Kobe (b) accelerograms

4 Triple Friction Pendulum: Does It Improve the Isolation Performance?

5 restrainers triggered (triple and double FPs)

4

4.5

sliding regime V triggered

3

sliding regime V not triggered 0.85

3.2

2

1

Northridge 5.5

4

triple FP double FP, μ1=μ2=4.5% double FP, μ1=μ2=6.5% without isolator (max=6.107m/s2)

3.5

sliding regime V triggered

3 2.5 2 1.5 1

sliding regime V not triggered

0.5 0

0

0.5

1

1.5

2

2.5

3

3.5

0 0

restrainers triggered (triple and double FPs)

6.45

triple FP double FP, μ1=μ2=4.5% double FP, μ1=μ2=6.5% without isolator (max=6.936m/s2)

5

max( | üs+üg | ) (m/s2)

b

1.12

Loma Prieta 6

max( | üs+üg | ) (m/s2)

a

31

1

2

PGA (m/s2)

3

4

5

6

7

PGA (m/s2)

Fig. 4.3 Isolation performances of triple FP according to literature and conventional double FP tuned by trial and error for Loma Prieta (a) and Northridge (b) accelerograms

As Fig. 4.2a shows also two triple FPs with slightly different friction tunings of the main sliding surfaces 1 and 4 are computed. These simulations show that increased friction on the main sliding surfaces 1 and 4 avoids that the triple FP is operated in sliding regime V which improves the isolation performance compared to the triple FP according to literature [1, 2]. Nevertheless, these two triple FPs also perform worse compared to the double FP.

4.6 Isolation Performance of Optimized Triple Friction Pendulum 4.6.1 PGAs of Optimization Due to the nonlinearity of friction damping of both FPs their isolation performance depends on their relative motion amplitude [5] and consequently on PGA. Therefore, the optimizations of both FPs must be done for selected PGAs. The following PGAs for optimization are chosen here: • PGAopt D 2.5 m/s2 representing a small DBE, • PGAopt D 5 m/s2 representing a medium to large DBE, and • PGAopt D 7.5 m/s2 representing a large DBE.

4.6.2 Isolation Time Period The optimizations of both FPs are made for the typical situation where the fundamental time period of the non-isolated structure is 0.83 s (1.2 Hz), the vertical load W is 12 MN,  s D 1% and the selected isolation time period is 3.5 s in order to shift the isolated structure into the time period range of good isolation. This leads to effective radii Reff  1 D Reff  4 D 1.522 m for the triple FP and Reff  1 D Reff  2 D 1.522 m for the double FP.

4.6.3 Articulated Slider Assembly of Triple FP As the intention of the articulated slider assembly of triple FP is to trigger relative motion in the bearing at even very small PGAs the friction coefficients 2 D 3 are assumed to be 1.75% similar to the published data. According to the design

32

F. Weber et al.

philosophy of the triple FP the effective radii Reff  2 D Reff  3 are designed to be smaller than Reff  1 D Reff  4 so that sliding regime II is activated without force step due to triggered restrainers on concave plates 1 and 3. In order to comply with this design methodology, Reff  2 D Reff  3 are selected to be eight times smaller than Reff  1 D Reff  4 as Reff  1 /Reff  2 D 8 is in accordance to the design given in [1, 2]. The displacement capacities d2 D d3 are scaled by approx. the same factor as the effective radii which yields d2 D d3 D 0.07 m.

4.6.4 Restrainers 1 and 4 The previous study demonstrated that the activation of sliding regime V due to restrainers 1 and 4 worsens the isolation of the structure dramatically. The previous study further demonstrated that the isolation deteriorates completely when the full displacement capacity of the bearing is used, i.e. all restrainers are triggered. Therefore, restrainers 1 and 4 are omitted for the optimization of the triple FP.

4.6.5 Optimization Parameters Due to the definition of the isolation time period of 3.5 s whereby Reff  1 and Reff  4 of the triple FP are given, due to the design of the articulated slider assembly of the triple FP which is in agreement with the common design philosophy of the triple FP and due to the neglect of restrainers 1 and 4 of the triple FP in order to avoid bad isolation of the structure due to sliding regime V and when all restrainers are triggered the triple FP can be optimized for minimum acceleration response of the structure (2) by variation of the friction coefficients 1 and 4 . Similarly, the double FP with same isolation time period as the triple FP and without end stoppers on sliding surfaces 1 and 2 is optimized for minimum acceleration response of the structure (2) by variation of the friction coefficients 1 and 2 . As 1 and 2 can be different a double FP with articulated slider is assumed.

4.6.6 Optimization Results Exemplarily for all optimization cases, i.e. four earthquakes scaled to three PGAopt , two optimization results are plotted in Figs. 4.4 and 4.5. The following can be observed: • The optimal friction coefficients of the triple and double FPs and the resulting absolute structural acceleration are very similar. • The optimal friction coefficients can be different (El Centro NS, Fig. 4.4) or equal (Loma Prieta, Fig. 4.5); if they are different it is not relevant which one is higher (valid for triple and double FPs).

4.6.7 Isolation Performance of Optimized Triple FP The isolation performances of the optimized triple and double FPs are computed for the El Centro NS, the Kobe, the Loma Prieta and the Northridge accelerograms that are scaled to PGAs ranging from 0.25 m/s2 up to 150% of PGAopt assuming that the FPs are optimized at the PGA corresponding to DBE and that the PGA of MCE is around 150% of the PGA of DBE. The isolation performances accompanied by the total bearing displacement as an important economical parameter are depicted in Figs. 4.6, 4.7, 4.8, and 4.9 for PGAopt D 5 m/s2 ; showing also the results due to the optimizations at 2.5 and 7.5 m/s2 is not possible due to the limited number of pages of the manuscript but the results are qualitatively similar. It is seen that the optimized triple and double FPs approximately yield the same absolute structural accelerations and total bearing displacements. This outcome is explained by the fact that the triple FP mutates to a double FP by the optimization which

4 Triple Friction Pendulum: Does It Improve the Isolation Performance?

a

33

b

triple FP, Tiso=3.5s, El Centro NS, PGAopt=5.0m/s2

double FP, Tiso=3.5s, El Centro NS, PGAopt=5.0m/s2 optimum

1.6 1.4 1.2 1 0.8 15

15

max( | üs+üg | ) (m/s2)

max( | üs+üg | ) (m/s2)

optimum

1.6 1.4 1.2 1 0.8 15

15

10

10 5 friction coefficient μ4 (%)

0

0

10

10 5

5

5 friction coefficient μ1 (%)

friction coefficient μ2 (%)

0

friction coefficient μ1 (%)

0

Fig. 4.4 Optimization results for triple (a) and double (b) FPs for the El Centro NS accelerogram scaled to 5 m/s2

a

b

triple FP, Tiso=3.5s, Loma Prieta, PGAopt=7.5m/s2

double FP, Tiso=3.5s, Loma Prieta, PGAopt=7.5m/s2 optimum

4 3 0

0 5

5 10

10 15

15 20

friction coefficient μ4 (%)

25 25

20 friction coefficient μ1 (%)

max( | üs+üg | ) (m/s2)

max( | üs+üg | ) (m/s2)

optimum

4 3 0

0 5

5 10

10 15

15 20

friction coefficient μ2 (%)

20 25 25

friction coefficient μ1 (%)

Fig. 4.5 Optimization results for triple (a) and double (b) FPs for the Loma Prieta accelerogram scaled to 7.5 m/s2

means that 1 and 4 must be designed similar to 1 and 2 of the optimal double FP and that the articulated slider assembly is not beneficial for maximum isolation due to its high stiffness which is equivalent to a low isolation time period. From Fig. 4.6 it is also evident that the conventional design of the triple FP [1, 2] is far from optimal and that the suboptimal solutions with equal friction coefficients perform almost as well as their optimal counterparts.

4.7 Summary First, the isolation performance in terms of absolute structural acceleration of the triple friction pendulum (FP) with parameters from literature is compared to that of the conventional double FP. Second, the triple and double FPs are optimized for minimum absolute structural acceleration and their isolation performances are compared subsequently. The first study demonstrates that the triple FP designed according to literature performs worse than the conventional double FP with suboptimal friction tuning. The second investigation demonstrates that both optimized FPs generate similar structural acceleration responses and total bearing displacements due to the fact that the triple FP mutates to the double FP due to the optimization.

34

1.2 1

PGAopt

0.8 0.6 0.4

conventional triple FP (μ1=3.1%, μ4=11.4%) conventional triple FP (μ1=5%, μ4=9%)

0.2 0 0

1

2

3

4

5

6

Tiso=3.5s, El Centro NS, PGAopt=5m/s2 0.45

0.35 0.3 0.25 0.2 0.15 0.1 0.05 0

7 7.5

optimal triple FP (μ1=3.5%, μ4=5.5%) optimal double FP (μ1=3%, μ2=5%) suboptimal triple FP (μ1=μ4=4%) suboptimal double FP (μ1=μ2=4%) conventional triple FP (μ1=3.1%, μ4=11.4%) conventional triple FP (μ1=5%, μ4=9%)

0.4

1.5 PGAopt

1.4

maxima of total bearing displacement (m)

optimal triple FP (μ1=3.5%, μ4=5.5%) optimal double FP (μ1=3%, μ2=5%) suboptimal triple FP (μ1=μ4=4%) suboptimal double FP (μ1=μ2=4%)

1.6

max( | üs+üg | ) (m/s2)

b

PGAopt

Tiso=3.5s, El Centro NS, PGAopt=5m/s2 1.8

1.5 PGAopt

a

F. Weber et al.

0

1

2

PGA (m/s2)

3

4

5

6

7 7.5

PGA (m/s2)

Fig. 4.6 Isolation performance (a) and total bearing displacement (b) of triple and double FPs optimized for El Centro NS accelerogram scaled to 5 m/s2

0.45

1.2 1

0.6

1.5 PGAopt

0.8

PGAopt

max( | üs+üg | ) (m/s2)

1.4

0.4 0.2 0

0

1

2

3 4 PGA (m/s2)

5

6

7 7.5

optimal triple FP (μ1=2%, μ4=6%) optimal double FP (μ1=0.5%, μ2=4.5%)

0.4 0.35 0.3 0.25 0.2 0.15

PGAopt

optimal triple FP (μ1=2%, μ4=6%) optimal double FP (μ1=0.5%, μ2=4.5%)

1.6

Tiso=3.5s, Kobe, PGAopt=5m/s2

b maxima of total bearing displacement (m)

1.8

1.5 PGAopt

Tiso=3.5s, Kobe, PGAopt=5m/s2

a

0.1 0.05 0

0

1

2

3 4 PGA (m/s2)

5

6

7 7.5

Fig. 4.7 Isolation performance (a) and total bearing displacement (b) of triple and double FPs optimized for Kobe accelerogram scaled to 5 m/s2

4 Triple Friction Pendulum: Does It Improve the Isolation Performance?

2 1.5 1 PGAopt

0.5 0

0

1

2

3 4 PGA (m/s2)

5

6

optimal triple FP (μ1=8%, μ4=8%) optimal double FP (μ1=10.5%, μ2=10.5%)

0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0

7 7.5

Tiso=3.5s, Loma Prieta, PGAopt=5m/s2

0.9

1.5 PGAopt

2.5

maxima of total bearing displacement (m)

optimal triple FP (μ1=8%, μ4=8%) optimal double FP (μ1=10.5%, μ2=10.5%)

3 max( | üs+üg | ) (m/s2)

b

PGAopt

Tiso=3.5s, Loma Prieta, PGAopt=5m/s2 3.5

1.5 PGAopt

a

35

0

1

2

3 4 PGA (m/s2)

5

6

7 7.5

Fig. 4.8 Isolation performance (a) and total bearing displacement (b) of triple and double FPs optimized for Loma Prieta accelerogram scaled to 5 m/s2

optimal triple FP (μ1=0.5%, μ4=0.5%) optimal double FP (μ1=0.5%, μ2=0.5%)

0.8

0.6 0.5

0.3

1.5 PGAopt

0.4

PGAopt

max( | üs+üg | ) (m/s2)

0.7

0.2 0.1 0

0

1

2

3 4 PGA (m/s2)

5

6

7 7.5

Tiso=3.5s, Northridge, PGAopt=5m/s2

0.25

optimal triple FP (μ1=0.5%, μ4=0.5%) optimal double FP (μ1=0.5%, μ2=0.5%) 0.2

0.15

0.1

0.05

0

1.5 PGAopt

b

PGAopt

Tiso=3.5s, Northridge, PGAopt=5m/s2

0.9

maxima of total bearing displacement (m)

a

0

1

2

3 4 PGA (m/s2)

5

6

7 7.5

Fig. 4.9 Isolation performance (a) and total bearing displacement (b) of triple and double FPs optimized for Northridge accelerogram scaled to 5 m/s2

Acknowledgements The authors gratefully acknowledge the financial support of MAURER SE.

References 1. Fenz, D.M., Constantinou, M.C.: Spherical sliding isolation bearings with adaptive behavior: theory. Earthq. Eng. Struct. Dyn. 37, 163–183 (2008) 2. Fenz, D.M., Constantinou, M.C.: Spherical sliding isolation bearings with adaptive behavior: experimental verification. Earthq. Eng. Struct. Dyn. 37, 185–205 (2008) 3. Ruderman, M.: Presliding hysteresis damping of LuGre and Maxwell-slip friction models. Mechatronics. 30, 225–230 (2015) 4. Weber, F., Distl, H., Braun, C.: Dynamic characterization and isolation performance of triple friction pendulum. In: Proceedings of the 8th World Congress on Joints, Bearings and Seismic Systems for Concrete Structures (IJBRC), Atlanta (GA, USA), 25–29 September 2016 5. Weber, F., Boston, C.: Energy based optimization of viscous-friction dampers on cables. Smart Mater. Struct. 19, 045025 (11pp) (2010)

Chapter 5

Experimental Investigation of the Dynamic Characteristics of a Glass-FRP Suspension Footbridge Xiaojun Wei, Justin Russell, Stana Živanovi´c, and J. Toby Mottram

Abstract Due to high strength- and stiffness-to-weight ratios, good durability performance in a variety of environments and quick installation, fibre reinforced polymers have increasingly been utilised for construction of highway and pedestrian bridges. Their relatively low mass and stiffness make these bridges potentially susceptible to vibration serviceability problems, which are increasingly governing the design. Currently, a lack of experimental data on the dynamic characteristics of polymeric composite structures is hindering their wider application and the development of design guidance. To fully exploit the benefits of using these structural materials in bridge engineering requires a better understanding of their dynamic behaviour. The aim of this paper is to utilise ambient vibration measurements to experimentally identify the dynamic characteristics (i.e., natural frequency, damping ratio and mode shape) of a glass fibre reinforced polymer deck suspension footbridge in the UK. It is found that the Wilcott footbridge possesses a relatively high density of vibration modes in the low frequency range up to 5 Hz and has damping ratios of most of these modes >1%. Keywords FRP suspension footbridge • Dynamic characteristics • Ambient vibration testing • Peak picking method • Stochastic subspace identification method

5.1 Introduction Since the first Fibre Reinforced Polymer (FRP) road bridge was built in Miyun, China in 1982 [1], thousands of bridges with FRP components have been built around the World [2]. The driver for this practice is this structural materials’ favourable properties, including: high strength- and stiffness-to-weight ratios, good durability and short installation time. FRP construction is typically employed for short-to-medium span bridges. The longest span to date is 63 m, achieved with the Aberfeldy footbridge that was constructed in Scotland in 1992 [3], using the same construction system as for the Wilcott footbridge. The benefits of using FRP as the structural material would be more prominent if longer bridge spans could be executed. Due to the lightweight and relatively low stiffness of glass FRPs, FRP bridges may be very lively and potentially suffer excessive vibration, causing user discomfort and affecting the bond in joints and between surfacing and the FRP superstructure [4]. Vibration serviceability is increasingly found to govern the design of FRP structures. A sound knowledge and understanding of their dynamic characteristics is therefore important for us having robust serviceability design procedures. Owing to a lack of experimental data on the dynamic characteristics we find that existing design guidance used in conventional material designs is usually employed. This approach may produce a conservative solution and compromise the benefits of using FRP in the first place. For example, in the AASHTO Guide Specifications for Design of FRP Pedestrian Bridges (1st Ed.) [5], a damping ratio of 2–5% is recommended. A damping ratio of at least 1% is expected to be present in every FRP structures, whilst such a high value as 5% is rarely achieved with pedestrian bridges made of other structural materials, such as steel, steel-concrete composite, aluminium and, even, concrete. In Prospect for New Guidance in the Design of FRP [6] the damping ratio is specified to be 1.5% for a conservative estimate in design analysis, and higher damping values may be used if these have been substantiated by representative experimental data. Given that the above recommendations for damping ratios are not necessarily based on extensive and comprehensive data from as-built FRP

X. Wei () • J. Russell • S. Živanovi´c • J.T. Mottram School of Engineering, University of Warwick, Coventry, CV4 7AL, UK e-mail: [email protected] © The Society for Experimental Mechanics, Inc. 2017 J. Caicedo, S. Pakzad (eds.), Dynamics of Civil Structures, Volume 2, Conference Proceedings of the Society for Experimental Mechanics Series, DOI 10.1007/978-3-319-54777-0_5

37

38

X. Wei et al.

Fig. 5.1 The Wilcott Bridge

structures there is a need to experimentally determine them. This paper contributes to this goal by providing a rare insight into the dynamic features of suspension footbridge near to the village of Nesscliffe, Shropshire, UK (Google Earth co-ordinates 52ı 450 54.8900 N and 2ı 550 04.6800 W). The paper starts with a description of the FRP bridge in Sect. 5.2, followed by a description of the ambient-based modal testing in Sect. 5.3. Identified modal parameters are reported in Sect. 5.4 with conclusions from their evaluation made in Sect. 5.5.

5.2 Bridge Description The Wilcott Bridge, shown in Fig. 5.1, is a single span suspension footbridge over the Nesscliffe A5 bypass road. The bridge has a width of 2.1 m and a span of 51.3 m. It consists of a lightweight glass FRP deck, two pairs of steel pylons, two steel cables with a span of 57.89 m, and four steel backstays and 20 steel hangers (10 per side). The FRP deck is constructed from pultruded components of E-glass fibre embedded in an isophthalic polyester matrix, which are adhesively bonded together to form an integral box section. The pultruded construction system is known today as Composolite, and is supplied by the American pultruder Strongwell. Ballast is employed to increase the mass of the deck. The deck was built in three units of approximately equal length that are connected by bonded interlocking splice joints and was integrally connected to the foundation, without the need for movement joints.

5.3 Ambient Vibration Testing To provide a precise and reliable characterization for the dynamic characteristics of the Wilcott footbridge an ambient testing programme was conducted by the authors from the University of Warwick, UK, on 09 September 2016. Acceleration responses of the deck were measured under the natural excitation of wind and road traffic passing underneath the bridge on a main truck road. During data recording the bridge was closed to pedestrian traffic. To identify the first few modes of the bridge we used the measurement grid in Fig. 5.2, having 26 measurement stations on top of the deck. These measurement locations included anchoring points of each hanger, the mid-span points and supports at the two span ends. Vertical accelerations only were recorded on the North side, whilst both vertical and horizontal accelerations were acquired on the South side. Figure 5.3 shows the accelerometers at a location on the South side. The measurement campaign was divided into seven set-ups. In each set-up we had eight Honeywell accelerometers QA750 (of nominal sensitivity of 1300 mV/g). A lateral accelerometer and a vertical accelerometer at measurement station 55 were used as reference vibration sensors, whilst the other six accelerometers are the roving transducers.

5 Experimental Investigation of the Dynamic Characteristics of a Glass-FRP Suspension Footbridge

39

South 52

53

54

55

56

57

58

59

60

61

62

63

2.1

51

Nesscliffe Side 1

2 4.21

3 4.64

4 4.65

5 4.55

6 4.75

7 2.32 2.32

8

9 4.65

10 4.65

11 4.65

12 4.65

Wilcott Side

13 4.21

50.24 Measurement station

North

Unit: m

Reference station

Fig. 5.2 Measurement grid on the deck

Fig. 5.3 Vertical and lateral accelerometers on the deck

The test programme was started from the Nesscliffe side of the bridge and progressed to the Wilcott side. The duration for each measurement set-up was 20 min. The sampling frequency was 256 Hz. Typical vertical and lateral acceleration time histories at the reference station (No. 55) in one of the seven set-ups are plotted in Fig. 5.4. The ambient vibration response level exhibited little variation between the seven different set-ups.

5.4 Modal Parameter Identification To identify the modal parameters of the Wilcott footbridge we employed the frequency-domain peak-picking method and the time-domain stochastic Sub-Space Identification (SSI) method. Whilst the peak-picking method was chosen to identify natural frequencies only, the SSI method was employed to identify natural frequencies, damping ratios and mode shapes.

5.4.1 Peak-Picking Method Since certain modes may not be observable in the power spectrum density at a particular measurement station, the Averaged Normalised Power Spectral Densities (ANPSD) [7, 8] of all the measurement stations can be used to identify all natural frequencies. The ANPSDs of the vertical and lateral accelerations are plotted in Figs. 5.5 and 5.6, respectively. The natural frequencies of the bridge are inferred from the peaks in the ANPSD curves. From the results in Fig. 5.5 there are six dominant peaks in the vertical direction, at frequencies of 0.94, 1.52, 2.22, 2.72, 3.20 and 3.89 Hz, and a less pronounced

40

X. Wei et al.

Vertical (m/s2)

0.3 0.15 0 −0.15 −0.3

0

200

400

600 Time (second)

800

1000

1200

200

400

600 Time (second)

800

1000

1200

Lateral (m/s2)

0.3 0.15 0 −0.15 −0.3 0

Fig. 5.4 Acceleration time histories measured at the reference station (No. 55) 100 0.94

10−2

1.52

2.22 2.72

ANPSD

3.20 3.89 4.17

10−4

10−6 0

1

2 3 Frequency (Hz)

4

5

Fig. 5.5 ANPSD for vertical acceleration records

peak at 4.17 Hz. As for the lateral direction with results in Fig. 5.6, the three dominant peaks are observed at 1.08, 1.55 and 4.19 Hz, with two smaller peaks at 2.19 and 3.20 Hz. In addition, there are two close frequency peaks at 2.47 and 2.58 Hz. It is seen that frequencies at certain peaks in the vertical direction are close to those observed in the lateral direction. This finding shows that there is uncertainty in interpreting whether the close peaks in the two orthogonal directions correspond to distinct modes or that they represent the same mode having both lateral and vertical components. The SSI method presented in Sect. 5.4.2 is used to further analyse these.

5.4.2 Stochastic Subspace Identification A reference-based data-driven stochastic SSI algorithm, built in the Matlab toolbox MACEC 3.2 (by K.U.LEUVEN Research & Development [9–12]), was applied for data pre-processing and modal parameter identification. The measured data were first de-trended, such that the DC components were removed. Then the data were filtered using a low-pass filter with a cut-off frequency of 20.48 Hz and resampled at 25.60 Hz.

5 Experimental Investigation of the Dynamic Characteristics of a Glass-FRP Suspension Footbridge

41

100

1.55

ANPSD

10−2 1.08 2.47 2.58 2.19

10−4

10−6

4.19 3.20

0

1

2 3 Frequency (Hz)

4

5

1

2 3 Frequency (Hz)

4

5

Fig. 5.6 ANPSD for lateral acceleration records 200

Model order

150

100

50

0

Fig. 5.7 The stabilization diagram of one of the setups

The vertical and lateral accelerations from all the sensors were used for modal parameter identification, with the model order parameter set to 200. The two channels corresponding to the reference stations (No. 55) were chosen as the reference channels. The stabilisation criteria were set to 1% for frequency, 5% for damping, 1% for modal assurance criterion and 0.8 for the low bound of the modal phase collinearity. Shown in Fig. 5.7 is a stabilisation diagram from one of the data set-ups, with the power spectral density of all the signals superimposed. Only the frequency content, up to 5 Hz, is presented in the figure, since this is the most relevant frequency range for pedestrian structures. Stable poles are presented by big red circles. It can be seen that the stable poles for natural frequencies are clearly identified, except for the one around 2.70 Hz. There are in total nine dominant modes identified, including three lateral bending, five vertical bending and one torsional mode. These nine mode shapes are illustrated in Figs. 5.8, 5.9, 5.10, 5.11, 5.12, 5.13, 5.14, 5.15 and 5.16. It can be observed that a majority of the modes found by the peak-picking method are identified by the SSI method. The only discrepancy is that two modes appear from peak-picking in the frequency range of 2.40–2.70 Hz, whereas Fig. 5.7 for SSI shows that there is only one consistent stable pole around 2.70 Hz. Moreover, the mode shapes of the identified lateral bending modes in Figs. 5.9, 5.11 and 5.16 are not very smooth. These findings might be due to the fact that the vertical acceleration signals are stronger than those in the lateral direction, and consequently the noise in vertical accelerations might be spoiling or hiding some lateral bending dominated modes [13]. The identified modes are summarised in Table 5.1. There is a relatively high mode density in the frequency range 0–5 Hz. From the first harmonic dynamic force generated from pedestrian walking the two vertical bending modes at the natural frequencies of 1.51 and 2.21 Hz and a lateral mode at 1.08 Hz are potentially excitable. The damping ratios of all vertical bending modes in Table 5.1 are not 1%, except for

42

X. Wei et al.

(a)

50 40

1

30

z

0

20 10

x

y

y

(b) 10

20

30

40

50

30

40

50

x

(c) z

1 0 10

20 x

Fig. 5.8 The first vertical bending dominated mode (f D 0.96 Hz,  D 2.49%)

50

z

(a)

40 30

0.5 0 1

20 10

x

0

y

y

(b)

1 0 10

20

30

40

50

x

z

(c) 0.5 0 10

20

30

40

50

x

Fig. 5.9 The first lateral bending dominated mode (f D 1.08 Hz,  D 0.31%)

the first of these modes, which has an exceptionally low damping ratio of 0.31%. Despite the potential for a number of modes being excited by pedestrians we find the bridge’s vibration performance to be satisfactory. Contributing to this good serviceability performance is the reasonable level of damping in almost every mode and the potential interaction between simultaneously excited modes. The vibration response of the bridge to a range of human actions is going to be subject to new work by the University of Warwick team.

5 Experimental Investigation of the Dynamic Characteristics of a Glass-FRP Suspension Footbridge

43

50 40

(a)

30

z

1 20

0 10

x

y

y

(b) 10

20

30

40

50

x

(c) z

1 0 10

20

30

40

50

x

Fig. 5.10 The second vertical bending dominated mode (f D 1.51 Hz,  D 1.91%)

50 40

z

(a)

30

0.5 0

20 10

1 y 0

x

(b) y

1 0 10

20

30

40

50

40

50

x

z

(c) 0.5 0 10

20

30 x

Fig. 5.11 The second lateral bending dominated mode (f D 1.56 Hz,  D 1.67%)

5.5 Conclusions Ambient testing of a suspension pedestrian bridge made with fibre reinforced polymer components was conducted to identify modal properties. The method of peak picking of the averaged normalised power spectral densities in the frequency-domain was employed to identify the natural frequencies, whilst the stochastic subspace identification method, in the time-domain, was employed to the identify natural frequencies, damping ratios and mode shapes. The combined use of the two independent analysis techniques led to identifying of as many as nine dominant vibration modes in the low frequency range of 0–5 Hz.

44

X. Wei et al.

50 40

(a)

30

1 z

20 10 x

y

y

(b) 10

20

30

40

50

30

40

50

x

(c) z

1 10

20 x

Fig. 5.12 The third vertical bending dominated mode (f D 2.21 Hz,  D 0.96%)

(a)

50 40

z

1

30

0

20 10

y

x

y

(b) 10

20

30

40

50

x

(c) z

1 0 10

20

30

40

50

x

Fig. 5.13 The fourth vertical bending dominated mode (f D 2.71 Hz,  D 1.89%)

These nine modes include three lateral bending, five vertical bending and one torsional. The damping ratios of most modes were found to be >1%, and this provided evidence to justify a recommendation in vibration serviceability design to have the damping ratio at 1% or higher. Despite some vibration modes having natural frequencies excitable in resonance by human walking actions the relatively high damping in the Wilmott footbridge ensures satisfactory serviceability performance.

5 Experimental Investigation of the Dynamic Characteristics of a Glass-FRP Suspension Footbridge

45

50

(a) 40

1 30 z

0

20 10

x

y

y

(b)

10

20

30

40

50

x

(c) 1 z

0 10

20

30

40

50

x

Fig. 5.14 The first torsional mode (f D 3.22 Hz,  D 0.82%)

50

(a)

40 30

z

1 20

0 10 y

x

y

(b) 10

20

30

40

50

x

(c) z

1 0 10

20

30

40

50

x

Fig. 5.15 The fifth vertical bending dominated mode (f D 3.86 Hz,  D 1.41%)

Acknowledgements This research work was supported by the UK Engineering and Physical Sciences Research Council [grant number EP/M021505/1: Characterising dynamic performance of fibre reinforced polymer structures for resilience and sustainability].

46

X. Wei et al.

z

(a)

50 40

0.5 0 −0.5 1 y

30 20 x

10

0

(b)

y

1 0 10

20

30

40

50

x

z

(c) 0.5 0 −0.5 10

20

30

40

50

x

Fig. 5.16 The third lateral bending dominated mode (f D 4.11 Hz,  D 1.34%) Table 5.1 Identified modal parameters of the Wilcott Bridge

SSI Frequency No. Mode description (Hz) 1 1st VD 0.96 2 1st LD 1.08 3 2nd VD 1.51 4 2nd LD 1.56 5 3rd VD 2.21 6 4th VD 2.71 7 1st T 3.22 8 5th VD 3.86 9 3rd LD 4.11

Damping (%) 2.49 0.31 1.91 1.67 0.96 1.89 0.82 1.41 1.34

Peak-picking Frequency (Hz) 0.94 1.08 1.52 1.56 2.22 2.72 3.20 3.89 4.19

VD vertical bending dominated mode, LD lateral bending dominated mode, T torsional mode

References 1. Ye, L.P., Feng, P., Zhang, K., Lin, L., Hong, W.H., Yue, Q.R.: FRP in civil engineering in China: research and applications. In: Proceedings of Fiber Reinforced Polymer: Reinforcement for Concrete Structures, pp. 1401–1412. Singapore (2003) 2. Wan, B.: Using fiber-reinforced polymer (FRP) composites in bridge construction and monitoring their performance: an overview. Advanced Composites in Bridge Construction and Repair, pp. 3–28 (2014) 3. Burgoyne, C., Head, P.: Aberfeldy Bridge–an advanced textile reinforced footbridge. In: Techtextil Syposium, pp. 1–9 (1993) 4. BD 49/01 Design Rules for Aerodynamic Effects on Bridges. In: Design Mannual for Roads and Bridges, vol. 1, Section 3, Part 17. ed: Highways England (2001) 5. AASHTO: Guide Specifications for Design of FRP Pedestrian Bridges, 1st edn. American Association of State Highway and Transportation Officials, Washington (2008) 6. Ascione, L., Caron, J.-F., Godonou, P., IJselmuijden, K.v., Knippers, J., Mottram, T., et al.: Prospect for New Guidance in the Design of FRP. EUR 27666 EN, 2016 7. Felber, A.J.: Development of a hybrid bridge evaluation system. PhD Thesis, The University of British Columbia (1994)

5 Experimental Investigation of the Dynamic Characteristics of a Glass-FRP Suspension Footbridge

47

8. Ren, W.-X., Peng, X.-L., Lin, Y.-Q.: Experimental and analytical studies on dynamic characteristics of a large span cable-stayed bridge. Eng. Struct. 27, 535–548 (2005) 9. Reynders, E., Schevenels, M., Roeck, G.D.: MACEC 3.2: A Matlab Toolbox for Experimental and Operational Modal Analysis. Department of Civil Engineering, KU Leuven (2014) 10. Peeters, B., De Roeck, G.: Reference-based stochastic subspace identification for output-only modal analysis. Mech. Syst. Signal Process. 13, 855–878 (1999) 11. Peeters, B., De Roeck, G.: Stochastic system identification for operational modal analysis: a review. J. Dyn. Syst. Meas. Control. 123, 659–667 (2001) 12. Reynders, E., Roeck, G.D.: Reference-based combined deterministic–stochastic subspace identification for experimental and operational modal analysis. Mech. Syst. Signal Process. 22, 617–637 (2008) 13. Brownjohn, J.M.W., Magalhaes, F., Caetano, E., Cunha, A.: Ambient vibration re-testing and operational modal analysis of the Humber Bridge. Eng. Struct. 32, 2003–2018 (2010)

Chapter 6

Vibration-Based Occupant Detection Using a Multiple-Model Approach Yves Reuland, Sai G. S. Pai, Slah Drira, and Ian F. C. Smith

Abstract Sensor-based occupant detection has the potential to make an important contribution to the development of structures of the future. Applications that may benefit from robust occupant detection include patient detection in hospitals, senior citizen housing facilities, personnel localization in emergencies as well as user behavior studies. In this contribution, an occupant detection and localization methodology based on recorded vibration time-series is outlined. The movement of an occupant on a floor generates vibrations that can be recorded by accelerometers. However, measured vibrations contain measurement noise and are contaminated by ambient sources of vibration such as machinery and nearby traffic. This contribution relies on using filtered vibration time-series to detect events of moving occupants and subsequently perform model-based localization of occupants using error-domain model falsification. The error-domain model-falsification methodology utilizes multiple models to deal with ambiguity related to the inverse problem of occupant localization. By explicitly incorporating uncertainty from various sources using engineering heuristics, error-domain model falsification provides a set of candidate locations based on measurements obtained through a coarse sensor configuration. The results from this methodology provide in a binary manner the presence or absence of an occupant and subsequently candidate locations of the occupant on the floor of a 200 m2 hall that is equipped with only four accelerometers. Keywords Human detection • Structural vibrations • Multiple model • Error-domain model falsification • Robust localization

6.1 Introduction Recent advances in sensing and computing technology have made available reliable sensors at reduced cost and tools for processing this data are leading the way for development of structures of the future. Sensor-based human detection requires availability of data from sensors and robust methods for interpretation of this data in presence of ambiguity. Human detection using sensor data has the potential to increase the safety of building occupants, security and general knowledge of building use. For example, this technology may be applied for patient detection in hospitals and senior-citizen housing facilities, security systems for banks, personnel localization in emergencies, energy management and user-behavior studies. Fall detection of elderly and patients in healthcare facilities is particularly likely to benefit from this technology. Current state-of-the-art fall detectors include community alarms, wearable detectors and cameras, which are a potential impedance to functioning of patients and undermine their privacy [1, 2]. An alternative approach for human detection and localization relies on the use of vibration data from sensors. Unlike the use of cameras and to a lesser extent, motion sensors for occupant detection, retrieval of vibration data from sensors does not infringe on the privacy of building occupants. Localization of an occupant has been proposed by Schloemann et al. [3] using vibration measurements from 241 sensors. Lam et al. [4] proposed an occupant detection methodology taking into consideration structural behavior to determine thresholds on classifying vibrations from occupant activities. Pan et al. [5] presented a building occupancy estimation system wherein data from low-resolution vibration sensors was used for occupant detection. They were able to distinguish between occupant activity and noise from ambient vibrations leading to tracking of occupants entering a room, their position and number of occupants in a room. Mirshekari et al. [6] used high energy components of signals and time-difference-of arrival to accurately localize human presence. Floor vibration measurement has even been used for gender classification of occupants

Y. Reuland () • S.G.S. Pai • S. Drira • I.F.C. Smith Applied Computing and Mechanics Laboratory, Swiss Federal Institute of Technology (EPFL), Lausanne 1015, Switzerland e-mail: [email protected] © The Society for Experimental Mechanics, Inc. 2017 J. Caicedo, S. Pakzad (eds.), Dynamics of Civil Structures, Volume 2, Conference Proceedings of the Society for Experimental Mechanics Series, DOI 10.1007/978-3-319-54777-0_6

49

50

Y. Reuland et al.

based on vibration characteristics [7]. However, most of the research in this domain has focused on using a high-density sensor layout, which is unrealistic for monitoring systems of most smart-buildings and which might induce unnecessary costs. Current human detection and localization methodologies are model-free, meaning that they rely only on processing and analyzing measurement data. Coupling the measured response with a structural behavior model of the building has the potential to reduce the number of sensors necessary for human localization. However, predictions of the structural response for vibrations induced by human footsteps are prone to multiple sources of uncertainty including material properties, geometry and boundary conditions. Therefore, a model-based data-interpretation methodology that is robust in presence of uncertainty is needed to perform human localization using vibration measurements from a sparse sensor configuration. In this paper, a procedure for human detection and localization is presented using floor vibration data. First a methodology to detect the presence of an occupant is described. Once the presence of an occupant is established, localization is carried out using error-domain model falsification (EDMF) [8], which is a model-based data interpretation methodology that has already been employed successfully for improving knowledge of the behavior of bridges [9, 10], water-supply-networks [11] and wind around buildings [12].

6.2 Methodology 6.2.1 Human Detection Detection of human presence on a floor is achieved through on a two-step procedure [4]. In the first step, structural characteristics of the floor slab are derived from prior ambient vibration measurements. Two types of information are obtained from these baseline measurements, the dominant frequencies of the slab at the measured locations as well as the level of vibrations that characterize ambient conditions (due to factors such as outside traffic). With the ambient signal taken as a Gaussian white noise process, the signal statistics that are obtained from the initial measurements, mainly the standard deviation of the signal, are calculated. With knowledge of the ambient vibration level, thresholds are set and then used to detect “anomalies” such as human steps. As a human step can be conceptualized to be an impact load, measurable levels of energy are carried by the shockwaves that are close to the dominant frequencies of the structure. Based on this observation, and in order to improve the signal-tonoise ratio for human detection, the recorded signals are bandpass-filtered around the dominant frequencies. A classic sixthorder Butterworth filter is used to filter the signals. Various recordings of several minutes taken on several days are then used to compute detection thresholds based on the standard deviation of the bandpass-filtered ambient-vibration accelerations. Human detection is achieved when two consecutively measured samples exceed the detection threshold.

6.2.2 Error-Domain Model Falsification Popper [13] asserted that data cannot be used to validate correct models, rather to falsify wrong models. Motivated through this assertion, error-domain model falsification was proposed by Goulet et al. [8] as a multiple model methodology, where wrong models are falsified using thresholds, which are defined by the uncertainties associated with the system. The EDMF methodology has been applied to fourteen full-scale systems [14]. In this paper, the application of this methodology is extended to human localization using acceleration data. Assume g() is a model with n identification parameters  i . The response of such a physics-based model is subjected to uncertainty from many sources such as model fidelity, geometric simplifications, material property and boundary conditions. Many of these uncertainty sources cannot be approximated using zero-mean independent Gaussian distributions. Moreover, some of these uncertainties are systematic in nature with unknown correlation between measurement locations. Let the combination of all these uncertainties at a measurement location i be "mod,i . If m such measurement locations are present, then y D [y1 , y2 : : : ym ] T is the vector of measured structural responses at m locations. The uncertainty associated with each measurement is "meas,i . If Qi is the true (unknown) structural response at a measurement location then it can be represented as the difference between model prediction and model error or the difference between measurement and measurement error as shown in Eq. (6.1)   Qi D gi    "mod;i D yi  "meas;i

(6.1)

6 Vibration-Based Occupant Detection Using a Multiple-Model Approach

51

where,  * is the vector of model parameter values and gi ( * ) is the predicted structural response at measurement location i. By rearranging Eq. (6.1), the residual between the model prediction and measurements is equal to the combined uncertainty at a measurement location, "c,i . In a probabilistic approach, these errors "mod,i , "meas,i and "c,i are represented as random variables Umod,i , Umeas,i and Uc,i , respectively. Thresholds are defined using the combined uncertainty Uc,i . For a target reliability of identification  2 f0,1g, the falsification thresholds, Thigh,i and Tlow,i , are computed as shown in Eq. (6.2)



1=m

TZ high;i

fUc;i ."c;i / d"c;i

D

(6.2)

Tlow;i

where, fUc;i is the PDF of the combined uncertainty and  is the target reliability of identification. Due to small number of measurements available, the target reliability of identification is corrected using the term 1/m, called as Šidák correction for multiple hypotheses testing [15]. In EDMF, the user generates model response for multiple values of model parameters . Then the residual at each measurement location is calculated as the difference between model prediction and measurement. For all measurement locations, model instances whose residuals lie outside the thresholds are falsified, as represented by Eq. (6.3). 8i 2 f1; ::; mg

Tlow;i  gi ./  yi  Thigh;i

(6.3)

All model instances not falsified using Eq. (6.3) are accepted into the candidate model set [16–18]. Due to the lack of information of the true uncertainty distributions, all candidate models are treated as equally probable. The localization challenge is treated as an inverse problem where the primary parameter to be identified is the location of the load.

6.3 Test Setup The application of human detection and localization has been performed on a continuous reinforced concrete slab that forms the entrance hall of a building on the EPFL campus. The slab is 24 cm thick, which is typical for buildings in Northern Europe. In addition to the high stiffness of the slab, a dense network of structural and non-structural walls underneath the slab results in relatively short spans of the slab (see Fig. 6.1). Most structural walls are in reinforced concrete, while nonstructural walls are built in unreinforced masonry. In combination with the rigid slab, the coarse sensor network makes the tested setup an unconventionally complex one. Therefore, the results obtained demonstrate a lower limit of the efficiency of human detection and localization with a multiple model approach. Roctest Actimon-X1 sensors are used to record accelerations in all three directions. The sensors have a measurement range of 1.5 g and a resolution of 0.05 mg. The frequency response of the sensor ranges from 0 to 200 Hz and the maximum sampling rate is 2000 Hz. The sensors are attached underneath the slab using screws.

6.4 Results Prior information of the vibration characteristics of the building are obtained from ambient-vibration measurements performed on 4 days. As described above, the most important features that are obtained are the fundamental frequency of the structure and the baseline level of ambient vibrations. The fundamental frequency of the slab is 18 Hz, while at sensor location 4, more energy is present around 24 Hz. Based on the mean level of measured ambient vibrations, thresholds are calculated for detection of occupants. In order to increase the signal-to-noise ratio, and accounting for the measured fundamental frequencies, the measured signal is bandpass-filtered using a sixth order Butterworth filter between 16 and 25 Hz. Best results are obtained if thresholds for human detection are set to six standard deviations of the filtered ambient signals. Comparison with raw signal, wavelet transforms (Mexican hat wavelet) and short-term average over long-term average (STA/LTA) ratios showed that for this application, a comparatively simple technique such as bandpass filtering provides the best results for human detection.

52

Y. Reuland et al.

Hall boundaries (walls above slab) 5.70 m

Structural walls below slab

3

4

18.20 m

Non-structural walls below slab

5.90 m 1

2

Sensor Locations

Step Location Walking pattern for Human detection 13.90 m

Fig. 6.1 Schematic view of the tested slab with sensor locations and walking pattern

Acceleration (mg)

2

0

−2 0

6

12

18

24

30

36

42

48

Time (s) Filtered Signal (Bandpass 18-24 Hz)

Detection Thresholds

Fig. 6.2 Successful human detection from sensor 1

The feasibility of human detection is shown for a tested walking pattern (see Fig. 6.1) on the slab. For the filtered signal of sensor 1 (see Fig. 6.2), footsteps that are close enough to the sensor, and that are not damped out by a structural wall underneath the slab, can be successfully detected. This example shows that even for a stiff slab having short spans, an accelerometer is able to measure the impact of a human step (person of approximately 85 kg) in a radius of more than 4 m. It is also interesting to note that steps that are done in different directions do not have the same amplitude. This might be linked to path-length dependent gates and is a promising feature for user behavior studies. As shown in Fig. 6.3, all four sensors (see Fig. 6.1 for sensor locations) showed the single footstep that is subsequently used for localization. Sensor 1 is closest to the footstep and logically recorded the highest acceleration. The amplitude of filtered acceleration for sensor 2 is comparable due to the absence of walls supporting the slab between the two sensors. Sensors 3 and 4 show a slight exceedance of the triggering thresholds, which shows that steps can be detected even if supporting walls are between the step source and the sensor. However, this observation is not true for all the steps that were analyzed and therefore, the sensor configuration that is necessary for robust human detection may need to take into account the walls and other types of supports underneath floor slabs. Localizing the presence of a human walking on the floor is treated as an inverse problem. The recorded vibration data is used to update knowledge about location of the occupant on a slab using EDMF. In this application, the only model parameter that is intended to be identified using EDMF is the location of an occupant. For application of EDMF, a finite element model of the hall is developed using ANSYS. Using the location of an occupant as a variable, a unit load is applied along a grid on the model and the vertical displacement at sensor locations is recorded. The model response is subjected to uncertainties from many sources, such as modulus of elasticity of the concrete deck, locations and stiffness of boundary conditions. Also, uncertainties arise from processing the measured vibration data using numerical integration in order to obtain displacements, as well as from the weight of the occupant and dynamic amplification

6 Vibration-Based Occupant Detection Using a Multiple-Model Approach

53 Sensor 2

2

Acceleration (mg)

Acceleration (mg)

Sensor 1

0

−2 19.5

19.8

20.1

2

0

−2 19.5

20.4

Time (s) Sensor 3

20.4

Sensor 4 1 Acceleration (mg)

Acceleration (mg)

20.1

Time (s)

1 0.5 0 −0.5 −1 19.5

19.8

19.8

20.1

20.4

0.5 0 −0.5 −1 19.5

Time (s)

19.8

20.1

20.4

Time (s)

Fig. 6.3 Signals induced by one footstep (see Fig. 6.1) recorded on the 4 sensors Table 6.1 Uncertainty sources and distribution

Source Measurement uncertainty (%) Modelling uncertainty (%) Sensor 1 location uncertainty (mm) Sensor 2 location uncertainty (mm) Sensor 3 location uncertainty (mm) Sensor 4 location uncertainty (mm)

Distribution Min/Mean Max/Std dev Gaussian

0

3

Uniform

35

85

Uniform

0.0006

0.0006

Uniform

0.0006

0.0005

Uniform

0.0010

0.0010

Uniform

0.0004

0.0004

of structural response due to this loading. These uncertainty sources are synthesized as modelling uncertainty. In addition, measurement uncertainty that is provided by the sensor manufacturer is taken into account. Finally, the uncertainty related to the exact position of sensors on the slab are derived from a parametric analysis using the finite element model, by changing the sensor location by ˙250 mm. The estimated distribution of these three uncertainty sources is shown in Table 6.1. The uncertainties are subsequently combined using a Monte Carlo simulation to obtain the combined uncertainty distribution. Once the combined uncertainty distribution is obtained, the thresholds of the error-domain are computed using Eq. (6.2) for a target reliability of identification of 0.95. In EDMF, a model instance is falsified if the residual between model prediction and measurement lies outside the threshold bounds. In this application, the model prediction is the vertical displacement at sensor locations for a unit load applied at various locations on the grid multiplied by the weight of the occupant and a dynamic amplification value that is taken equal to 1.3. For measurement response, a single step of a human being detected (see Fig. 6.3) is considered. The maximum displacement experienced for a step, obtained through numerical integration using the trapezoidal integration scheme, is considered to be the measurement response. The difference between measured and predicted vertical displacement response is then calculated for all possible occupant locations. Figure 6.4 shows the residuals calculated for sensor location 1. Falsified location instances that fall between thresholds for sensor 1 are falsified by another sensor.

Y. Reuland et al.

Displacement at Sensor 1 (x10−6m)

54 2

0

−2

−4 0

400 Model instances

Falsification Thresholds

800

Measured Value

Candidate Models

Combined Uncertainty

Falsified Model Instances

Fig. 6.4 Falsification plot for sensor 1 based on the maximum displacement induced by a footstep signal

Sensor 1

Sensor 2

Sensor 3

Sensor 4

19.5 m

19.5 m

19.5 m

19.5 m

15

15

15

15

10

10

10

10

5

5

5

5

0

6.5

13.9 m

0

6.5

13.9 m

0

6.5

13.9 m

0

6.5

13.9 m

Accepted model instances (by sensor indicated above subplot) Falsified Model Instances Hall boundaries Sensor Locations Fig. 6.5 Instances of human locations that are falsified by each sensor

Similarly, locations of the occupant are falsified using measurement data from all four sensors as shown in Fig. 6.5. From the figure, it can be seen that sensor 1, which is closest to the true location of the person, falsifies most locations of the occupant and is most informative of all sensors. In EDMF, the final candidate location set is comprised of locations not falsified by any of the sensors. Therefore, combining the information from all four sensors, the candidate location set is obtained and shown in Fig. 6.6. This set of candidate locations includes the true location of the step taken by the occupant. From Fig. 6.6, it can be seen that there are locations not falsified at the periphery of the hall and between sensor locations indicating a more comprehensive study of the sensor layout configuration might be useful in improving the accuracy of identifying candidate locations. Entropy-based justification will help in determining the usefulness of additional sensors for falsifying locations at the periphery. A detailed study of the model class required for localization might be useful in improving the robustness of candidate occupant locations obtained. Structural parameters such as support stiffness and material properties such as density and Young’s modulus were treated as deterministic and known values.

6.5 Discussion In this paper, a methodology has been presented for human detection and localization that can explicitly take into consideration ambiguities of the inverse problem. The methodology utilizes error-domain model falsification for updating knowledge regarding the presence and location of an occupant in a hall using vibration data. The methodology is

6 Vibration-Based Occupant Detection Using a Multiple-Model Approach

55

19.5 m

15

Initial Model Set 10

Candidate Model Set Sensor Locations True Location Hall delimitations

5

0

6.5

13.9 m

Fig. 6.6 Candidate locations (green) for the human presence derived from EDMF and 4 sensors compared the true location of a person

demonstrated to provide a set of possible locations that include the true location of an occupant on a comparatively stiff slab using a sparse configuration of commercially available sensors. In order to improve model-based localization of human presence, a refined model that includes other parameters and their uncertainty and allows simulation of the dynamic response of the slab to human induced footsteps is needed. Also, an enhanced model would allow the engineer to select the optimal number and location of sensors to improve detection and localization of human presence. Finally, the automatic detection of footsteps and the localization of multiple consecutive steps will further decrease the uncertainty on human location since impossible walking patterns can be falsified. Acknowledgments This work was funded by the Swiss National Science Foundation under Contract No. 200020_169026.

References 1. Alwan, M., Rajendran, P.J., Kell, S., Mack, D., Dalal, S., Wolfe, M., Felder, R.: A smart and passive floor-vibration based fall detector for elderly. In: 2006 2nd International Conference on Information & Communication Technologies, IEEE, pp. 1003–1007 (2006) 2. Yu, X.: Approaches and principles of fall detection for elderly and patient. In: 10th International Conference on e-health Networking, Applications and Services, 2008. Health Com 2008, IEEE, pp. 42–47 (2008) 3. Schloemann, J., Malladi, V.S., Woolard, A.G., Hamilton, J.M., Buehrer, R.M., Tarazaga, P.A.: Vibration event localization in an instrumented building. In: Experimental Techniques, Rotating Machinery, and Acoustics, vol. 8, pp. 265–271. Springer (2015) 4. Lam, M., Mirshekari, M., Pan, S., Zhang, P., Noh, H.Y.: Robust occupant detection through step-induced floor vibration by incorporating structural characteristics. In: Dynamics of Coupled Structures, vol. 4, pp. 357–367. Springer (2016) 5. Pan, S., Bonde, A., Jing, J., Zhang, L., Zhang, P., Noh, H.Y.: Boes: building occupancy estimation system using sparse ambient vibration monitoring. In: SPIE Smart Structures and MaterialsC Nondestructive Evaluation and Health Monitoring, pp. 90611O–90611O. International Society for Optics and Photonics (2014) 6. Mirshekari, M., Pan, S., Zhang, P., Noh, H.Y.: Characterizing wave propagation to improve indoor step-level person localization using floor vibration. In: SPIE Smart Structures and MaterialsC Nondestructive Evaluation and Health Monitoring, pp. 980305–980305. International Society for Optics and Photonics (2016) 7. Bales, D., Tarazaga, P., Kasarda, M., Batra, D.: Gender classification using under floor vibration measurements. In: Allen, M., Mayes, R.L., Rixen, D (eds.) Dynamics of Coupled Structures, Conference Proceedings of the Society for Experimental Mechanics Series, vol. 4, pp. 377–383. Springer International Publishing (2016) 8. Goulet, J.-A., Coutu, S., Smith, I.F.C.: Model falsification diagnosis and sensor placement for leak detection in pressurized pipe networks. Adv. Eng. Inform.. Elsevier 27(2), 261–269 (2013) 9. Pasquier, R., Goulet, J.-A., Acevedo, C., Smith, I.F.C.: Improving fatigue evaluations of structures using in-service behavior measurement data. J. Bridge Eng.. American Society of Civil Engineers 19(11), 4014045 (2014) 10. Pasquier, R., Angelo, L.D., Goulet, J.-A., Acevedo, C., Nussbaumer, A., Smith, I.F.C.: Measurement, data interpretation, and uncertainty propagation for fatigue assessments of structures. J. Bridge Eng.. American Society of Civil Engineers 21(5), 04015087 (2016) 11. Moser, G., Paal, S.G., Smith, I.F.: Performance comparison of reduced models for leak detection in water distribution networks. Adv. Eng. Inform. 29, 714–726 (2015) 12. Vernay, D.G., Raphael, B., Smith, I.F.C.: A model-based data-interpretation framework for improving wind predictions around buildings. J. Wind Eng. Ind. Aerodyn.. Elsevier 145, 219–228 (2015)

56

Y. Reuland et al.

13. Popper, K.: The Logic of Scientific Discovery. Routledge (1959) 14. Smith, I.F.C.: Studies of sensor-data interpretation for asset management of the built environment. Front. Built Environ.. Frontiers 2, 8 (2016) 15. Šidák, Z.: Rectangular confidence regions for the means of multivariate normal distributions. J. Am. Stat. Assoc.. Taylor & Francis Group 62(318), 626–633 (1967) 16. Goulet, J.-A., Kripakaran, P., Smith, I.F.C.: Multimodel structural performance monitoring. J. Struct. Eng.. American Society of Civil Engineers 136(10), 1309–1318 (2010) 17. Goulet, J.-A., Smith, I.F.C.: Structural identification with systematic errors and unknown uncertainty dependencies. Comput. Struct.. Elsevier 128, 251–258 (2013) 18. Goulet, J.-A., Michel, C., Smith, I.F.C.: Hybrid probabilities and error-domain structural identification using ambient vibration monitoring. Mech. Syst. Signal Process. 37(1), 199–212 (2013)

Chapter 7

Vibration Assessment and Control in Technical Facilities Using an Integrated Multidisciplinary Approach Nicholas Christie, James Hargreaves, Rob Harrison, and Francois Lancelot

Abstract Scientific research is advancing rapidly in several fields including materials science, applied physics, life sciences, bio and nanotechnology. This research often brings with it specialist imaging requirements to resolve ever finer details over time. The equipment needed to perform this imaging is often very sensitive to vibration in addition to factors including acoustic coupling, electromagnetic, thermal and airflow effects. This paper discusses the application of end-to-end vibration engineering to address these low vibration challenges. Using measurement data from the site and surrogate locations combined with advanced simulations allows a deeper understanding of the specific project implications and reduces risk and cost. This paper also illustrates how Arup deployed such an integrated multidisciplinary approach to assess and control vibration at the planning and design stages and contributed to the delivery of some of the most advanced scientific and industrial low vibration environment facilities in the world. Keywords Vibration • Technical facility • Laboratory • Assessment • Survey • Mitigation • Foundation • Ground • Structure • Passive control • Active control

7.1 Introduction Microscopy is one of the key aspects of scientific research and is continuously evolving to enable higher resolution powers to be achieved which then enables smaller detail sizes to be resolved. Electron based microscopy has enabled images to be produced for particle, then atomic and, in the future, sub-atomic matter with sub-Angstrom resolution considered to be the state-of-the-art. The performance of the microscope itself, however, depends on the environment within which it is located. This is typically an imaging suite in a university or other research building and while ideally this building would be located in a relatively quiet environment, in reality this is not necessarily possible. Environmental factors that can influence the microscope performance include structural vibration, acoustics, electromagnetic fields, thermal performance and air flow. In this paper the effects of structural vibration on electron and similar microscope performance will be presented and discussed in the context of a new building design. This will include criteria selection, vibration survey methods, modelling and analysis methods to guide the design and vibration mitigation to control or minimize the influence of structural vibration.

7.2 Structural Vibration Structural vibration is vibration that is transmitted through a building structure and eventually arrives at the floor plate structure of the imaging suite and then is transmitted to the microscope equipment. The sources of such vibration are many and varied but can be classified as internal or external. Internal vibration sources are those sources which lie within the

N. Christie • F. Lancelot () Arup North America Limited, Advanced Technology & Research, 560 Mission Street, Suite 700, San Francisco, CA, USA e-mail: [email protected]; [email protected] J. Hargreaves Ove Arup & Partners UK, Advanced Technology & Research, The Arup Campus, Solihull, UK e-mail: [email protected] R. Harrison Ove Arup & Partners UK, Advanced Technology & Research, 3 Piccadilly Place, Manchester, UK e-mail: [email protected] © The Society for Experimental Mechanics, Inc. 2017 J. Caicedo, S. Pakzad (eds.), Dynamics of Civil Structures, Volume 2, Conference Proceedings of the Society for Experimental Mechanics Series, DOI 10.1007/978-3-319-54777-0_7

57

58

N. Christie et al.

project boundary and include footfall, MEP plant, elevators as well as other equipment, machinery and in some cases test rigs. For a new building project, structural vibration from internal sources can be influenced and controlled to some degree by design or through facility management. External vibration sources are those sources which lie outside the project boundary and include surface and underground railway, highway, construction and demolition. Typically the vibration is transmitted from the sources through the ground to the new building foundation in the form of waves. These sources are not within the control of the project but the vibration transmitted from them to the new project must be somehow controlled and managed. Structural vibration has several characteristics which need to be considered in design: frequency, amplitude, wavelength, direction, temporal and spatial. The sources and transmission paths can be complex and the evaluation of vibration and its characteristics requires survey methods as well as modelling and analysis.

7.3 Criteria Criteria for vibration are essentially limits on vibration magnitude above which it is considered that the equipment or process performance might be impaired. Criteria can take the form of generic criteria published in [1, 2] and discussed in [3]. Generic criteria cover general classes of process and vibration sensitive equipment including electron microscopes and are relevant at the planning and even design stages where it may still be too early for specific manufacturer’s equipment to be identified with any confidence. The VC criteria are presented in [3] and shown in Fig. 7.1. The VC criteria have a number of aspects to consider. The VC vibration magnitudes are 100.0 m/s, which is also the average threshold of human vibration perception so the criteria are relevant to small vibrations which are generally subperceptible. The VC criteria are defined over a specific frequency range, namely 1–80 Hz. The format of the vibration is the RMS velocity in the 1/3rd octave band, although as discussed below there are a number of methods in which this can be evaluated. The criteria apply to vibration along the vertical or two horizontal axes. The criteria in [1] are applicable to “vibration measured on the building structure” and similarly in [2], however it is actually the vibration transmitted to the equipment that will be determinative of overall vibration performance. According to the above guidance, the criteria VC-D and VC-E might apply to electron and similar microscope equipment. Manufacturer’s criteria are intended to define limits

Fig. 7.1 VC criteria for vibration sensitive equipment

7 Vibration Assessment and Control in Technical Facilities Using an Integrated Multidisciplinary Approach

59

Fig. 7.2 Manufacturers’ criteria ranges for electron microscopy and VC criteria Instrument or equipment item (e.g. microscope, balance etc. Vibration mitigation (e.g. isolation, optical, active table)

Vibration at equipment support points = space vibration performance

Floor Vibration on slabe = structural vibration performance

Fig. 7.3 Structural vibration and space vibration performance

for vibration for a specific product or item of equipment. Such criteria might be defined with reference to the VC criteria or might be defined using an alternative format. A selection of manufacturer’s criteria is presented in Fig. 7.2 (SEM D Scanning Electron Microscope, TEM D Transmission Electron Microscope, STEM D Scanning Transmission Electron Microscope, STM D Scanning Tunnelling Microscope). In project work it is necessary to consult the manufacturer to confirm the format in which the criteria are defined. For project sites and buildings vibration generally varies with time leading to a need for some sort of metric, average and maximum to give the most common examples, to evaluate. Hence there are two key aspects of the criteria that need to be defined: the vibration limit and metric. The limit is simply the magnitude of vibration above which the equipment or process function starts to become impaired. The metric defines precisely how the temporal vibration must be evaluated before a comparison with the limit is made. Vibration in this context might be vibration measured during a survey or it might be a prediction from a computational model. Another key consideration with criteria is whether it is considered feasible, or even possible, to achieve them through structural design. Where some of the more demanding criteria such as VC-G, VC-F and VC-E might not realistically be achieved through structural design, it is necessary to introduce mitigation and to consider vibration transmission to the equipment rather than the structure. Enhancing the structural vibration performance with mitigation can produce a space vibration performance that is better than that of the structure and which does meet the criteria (see Fig. 7.3).

60

N. Christie et al.

7.4 Vibration Survey Vibration surveys enable the vibration performance of a site to be evaluated directly and are a key aspect of the development of technical building projects from the planning stage right through to commissioning. In this paper, attention is focused mainly on survey work at the planning stage where a proposed project site requires evaluation. The site is in an urban location with a main highway to the east and west of the site boundary and with a smaller road to the south [4]. This borehole is at a location that will be indicative of the future characterisation suite performance. The instrumentation at a typical measurement location is shown in Fig. 7.4 and comprises a steel canister (see inset image) placed at the base of a borehole in the ground and containing a tri-axial accelerometer arrangement connected to the data acquisition system. In this way vertical and horizontal ground vibration, at the building foundation elevation, is measured over a 24 h time period. The vertical vibration signal output from this test is shown in Fig. 7.5. The main source of vibration is highway traffic comprising cars, vans, trucks and buses and is generated by the interaction of vehicle dynamics with highway surface irregularity. The two most common metrics, namely average and maximum would be reported in most if not all projects. However, average would not typically be selected unless consultation with the user and manufacturer established that this would give a good indication of the vibration that might be disruptive in research practice. The maximum is by definition the worst case and needs to be evaluated and reported, however, in cases where there a very few instances in time when this occurs it can be less useful. The exceedence metric is a useful way of expressing temporal vibration performance and the 1% exceedence vibration is shown in Fig. 7.4 and indicates performance of VC-E compared to VC-D using maximum. 1% exceedence as a metric needs to be agreed with the project in the context of the equipment and methods being used but it typically means that VC-E would be exceeded for just a few minutes in each day and may be acceptable. In cases where image capture is not a process that can be quickly and easily repeated then several minutes of exceedence might be a concern.

Fig. 7.4 Borehole vibration survey

7 Vibration Assessment and Control in Technical Facilities Using an Integrated Multidisciplinary Approach

61

1.00E-03

VC-A

1.00E-04

VC-B VC-C

RMS Velocity (m/s)

VC-D VC-E

1.00E-05

VC-F VC-G Max max-hold: Vertical 1% exceedance: Vertical

1.00E-06

Linear ave: Vertical

1.00E-07

1.00E-08 1

10 rd 1/3 Octave Centre Frequency (Hz)

Fig. 7.5 Measured borehole vibration vs. VC-criteria

7.5 Computational Methods During the design stage of a building project the structural performance attributes need to be evaluated and the design steered to meet the respective criteria. For low vibration environment projects it is quite common for the strength design to be developed further to meet the vibration criteria with particular attention needed to the grid size, suspended slab thickness and the foundation structure. Computational methods such as finite element analysis (FEA) are a useful means of predicting structural vibration performance for the various dynamic loads associated with internal and external sources of vibration. Footfall is often the most critical dynamic loads for suspended slab structures, i.e. laboratory floors, and it is not normally possible to meet criteria such as VC-A with a strength design alone and without careful consideration of grid size and slab thickness. The effects of footfall on the structural design can be evaluated using published industry methods such as [5, 6]. An example of floor plate FEA using the GSA analysis software [7] is shown in Fig. 7.6. In this case the floor plate supports a typical laboratory space function with criteria VC-A. Footfalls in circulation areas, corridors and laboratory spaces need to be considered. As with test data processing, some consideration of the occurrence of peak events is necessary. Footfall induced vibration can produce resonant and impulsive response types. Resonant response involves a build-up of vibration with each footfall and occurs because one or more of the load harmonics align with a structural mode for a given walking speed. This occurs with so-called low frequency slab designs, with fundamental vertical natural frequencies of 10 Hz or less. High frequency floors exhibit impulsive response where each footfall produces a decaying impulse response. When comparing with the VC criteria it is necessary to evaluate the maximum RMS velocity in the 1/3rd octave band. A rough estimate of 1/3rd octave band RMS can be obtained by taking 70% of the whole signal RMS. However, this can be calculated by passing the predicted vibration time-history though the ANSI 1/3rd octave band filters [8] or by modal response based on modes within the 1/3rd octave bands. The predicted response of the floor plate in Fig. 7.6 is within VC-A for more than 90% of the floor plate area. The vibration response is higher than VC-A in some small areas and is lower in others. Floor plate areas around columns are clearly less mobile and in this case have a vibration performance of VC-B to VC-C or better. This makes it possible to realise more vibration performance out of a structure designed to the VC-A criteria and enables equipment that might require VC-B or VC-C to be used on the floor plate subject to feasibility constraints. FEA can be used to assess the vibration performance of foundation structures which is the part of the building structure where ground vibration transmitted from external sources such as railway and highway traffic, first arrives. An example is shown in which a ground bearing or raft foundation design has been proposed which requires a thickness of 350 mm to meet the strength performance requirements. The objective of the FEA study in this case is to determine whether this slab thickness would enable the vibration criteria to be met or whether it would need to be increased. A review of the principles of soil structure interaction in a low vibration environment context is available in [9]. The structural dynamics of the foundation

62

N. Christie et al.

Fig. 7.6 FEA analysis of footfall induced vibration on a suspended floor plate Concrete

All elements, 2D Axisymmetric elements

Foundation/Fill

Sandstone

Half-Space

z y

x

Fig. 7.7 FEA analysis of ground and foundation structure

depend on several parameters: slab mass, slab stiffness, slab structural damping, column take down loads, soil stiffness, soil hysteretic damping, radiation damping, soil stiffness, soil mass and soil-structure interaction. These parameters combined with the wave behaviour of the vibration in the ground result in relatively complex dynamic behaviour for which some simplification is necessary in order to apply FEA in a pragmatic way at the design stage. On this basis an axisymmetric model of the foundation structure and ground has been developed and used in conjunction with measured vibration data, see Fig. 7.7. In this case the axis of the model is aligned with the foundation pad centreline and a finite radius of foundation slab concrete is modelled. The take-down load of the columns is included as discrete mass centred on the model axis. The upper layers of soil are modelled discretely with the deep soil layers all simplified to an elastic half-space type continuum. Viscous dampers are applied to the vertical edge of the model in the horizontal and vertical directions. These dampers are known as Lysmer dampers and represent the radiation damping of the ground [10]. This model has been analysed using the MSC.NASTRAN finite element direct frequency response solver [11]. The external sources generate vibration in the ground which is transmitted to the foundation structure as an enforced motion. The first step in the analysis is to use the

7 Vibration Assessment and Control in Technical Facilities Using an Integrated Multidisciplinary Approach

63

7.000

VC-D

250-464 tonnes takedown 1.2m thick pad 1.35m radius pad 0.4m radius column

RMS Velocity (x1E-6 m/s)

6.000 5.000 4.000

VC-E 3.000 2.000 1.000 0.000 0.000

10.000

VC-D VC-E Test Data Max Load:- 350mm Max Load:- 800mm Max Load:- 1200mm Min Load:- 350mm Min Load:- 800mm

20.000

30.000

40.000

50.000

60.000

70.000

80.000

1/3 Octave Band Frequency (Hz) Min Load:- 1200mm

T/HIS

Fig. 7.8 Predicted foundation slab vibration using FEA model and measured data

reciprocity principle [12] to determine forces acting on the ground boundary that would produce the measured vibration on the foundation structure. This first step is necessary to enable a forced response analysis to be carried out which can enable the vibration response for various design assumptions to be evaluated, something that would not be possible with an enforced motion loading. An example of the output from this analysis is shown in Fig. 7.8 for the given structural configuration but assuming three slab thicknesses 350, 800 and 1200 mm and two different column take-down loads. In this case increasing the foundation thickness from 350 mm would be of marginal benefit, barely reducing the response by one full VC band from a baseline performance of VC-D, which meets the performance requirements. Where mitigation is being considered, as is the case here, this analysis needs to be combined with transmissibility data to estimate the slab vibration performance including mitigation. This is discussed in Sect. 7.6 below.

7.6 Mitigation Mitigation is a common aspect of technical facility design and arises because often structural vibration needs to be enhanced to meet the equipment criteria, especially where the more demanding criteria, such as VC-E and more onerous, are concerned. Mitigation might also be a more efficient means of achieving the required space performance or it may be more desirable, for example, where some amount of future-proofing is needed; not uncommon where research equipment is being continuously developed and a facility might be operational for 25 years or more. Mitigation systems can be grouped into passive or active types. Passive systems comprise relatively simple, unpowered systems which exploit basic physical principles. Common examples are low vibration bases and keel slabs. It should be noted that passive isolation systems are often designed into the equipment itself. Sometimes this is called internal isolation and care is needed when specifying keel slab systems to avoid “stacking” the isolation, i.e. where the internal and keel slab isolation frequencies are close or align. Active systems comprise devices that use computer controlled electro-mechanical actuators to deliver a dynamic force which minimises the vibration transmitted to the sensitive imaging and other equipment. The input to active systems is the floor vibration and the control system is designed to minimise this vibration using actuator arrays to deliver a platen vibration that meets the required criteria over a given frequency range, normally 1–80 Hz. One advantage of active control is that it is not based on a spring mass system and hence does not have an amplification region around the isolation frequency. This leads to better low frequency performance and also means it can work in the presence of internal isolation. The performance of both passive and active control systems needs to be specified and realised in design and manufacture. The vibration transmitted to the equipment from keel slab passive and two typical active control systems using the measured bore hole vibration can be estimated using theoretical transmissibility for each control system as shown in Fig. 7.9.

64

N. Christie et al. 1.00E-03

1.00E-04 VC-A VC-B VC-C

1.00E-05

VC-D

RMS Velocity (m/s)

VC-E VC-F

1.00E-06

VC-G 1% exceedance: Vertical 1% exceedance: Vertical: Keel slab 1% exceedance: Vertical: Active Isolation 1

1.00E-07

1% exceedance: Vertical: Active Isolation 2

1.00E-08

1.00E-09

1.00E-10 1

10 1/3rd Octave Centre Frequency (Hz)

Fig. 7.9 Predicted effect of active control system using measured bore hole vibration

7.7 Summary Successful technical facility design involves the management of internal and external sources of vibration from the start of the project, through the design stages and on to completion. The site vibration performance is one of the fundamental inputs to the vibration engineering aspects of the project and requires specific methods, equipment and results interpretation to enable advice of value to be provided. Computational methods are another important aspect of the design process which enables the structural vibration performance to be evaluated and compared with the criteria at the design stage. Examples have been presented of how FEA has been applied to footfall induced vibration on suspended slabs and, in conjunction with survey data, how a foundation design should be checked for vibration performance. An overview of passive and active vibration control has been provided and an illustration of how passive and active control can enhance structural vibration has been given.

References 1. 2015 ASHRAE Handbook, Heating Ventilating and Air-Conditioning Applications SI Edition 2. Institute of Environmental Sciences and Technology, IEST-RP-CC012.2 Considerations in Cleanroom Design (2007) 3. Amick, H., Gendreau, M., Busch, T., Gordon, C.: Evolving criteria for research facilities: I – vibration. Reprinted from SPIE Conference 5933: Buildings for Nanoscale Research and Beyond, San Diego, CA, 31 July 2005 to 1 August 2005 4. Manchester Engineering Campus Development (http://www.mecd.manchester.ac.uk/) 5. Willford, M., Young, P.: A Design Guide for Footfall Induced Vibration of Structures. Concrete Centre, December (2006) 6. Murray et al. Structural Design Guide – Section 11. American Institute of Steel Constructors (1997) 7. www.oasys-software.com/GSA 8. ANSI-S1.11: Specification for Octave-Band and Fractional-Band Digital Filters. American National Standards Institute 9. Amick, H., Xu, T., Gendreau, M.: The role of buildings and slabs-on-grade in the suppression of low-amplitude ambient ground vibrations. In: 11th International Conference on Soil Dynamics & Earthquake Engineering 10. Lysmer, J., Kulmeyer, R.L.: Finite dynamic model for infinite media. J. Eng. Mech. ASCE. 95, 859–875 (1969) 11. MSC.NASTRAN v2014 User Manual 12. Richart, F.E., Hall, J.R., Woods, R.D.: Vibration of Soils and Foundations. Prentice-Hall, Englewood Cliffs (1969)

Chapter 8

Iterative Pole-Zero Model Updating Using Multiple Frequency Response Functions M. Dorosti, R.H.B. Fey, M.F. Heertjes, M.M.J. van de Wal, and H. Nijmeijer

Abstract For accurate positioning of motion systems having an accurate yet low-order dynamic model from actuators to sensors and to unmeasured performance variables is crucial. A (reduced) Finite Element (FE) dynamic model may be a good candidate. However, a FE model in practice has limited accuracy in describing the dynamic behavior of the system for nano positioning performance. This can be either due to the simplifications in the FE modeling or due to system variations. To improve the dynamic properties of the (reduced) FE model toward dynamic properties of the real system, an Iterative Pole-Zero (IPZ) model updating procedure that updates poles and zeros of a single Frequency Response Function (FRF) was recently proposed. Using more FRFs from different actuator/sensor configurations helps to better improve the dynamic properties of the (reduced) FE model. In this paper, an extension of IPZ model updating to use multiple FRF measurements is presented. The proposed method is verified using a simulated experiment of a pinned-sliding beam structure. Keywords FE modeling • Model updating • Least square optimization • Unmeasured performance variables

8.1 Introduction Model updating techniques are well-known tools to improve the accuracy of a FE model. In principle there are two types of model updating techniques: direct methods [1] and iterative methods [2]. Within the iterative methods, there are two categories. The first category contains modal-based techniques concerned with updating modal properties such as poles, zeros, and mode shapes in an attempt to reduce the residuals between numerical and measured modal quantities [2–4]. Recently, an IPZ model updating technique is introduced that tries to update both the poles and the zeros of a single FRF of the system in an iterative manner. The second category attempts to reduce the residuals between numerical and measured FRFs directly [5, 6]. In model updating, it is sometimes clear which (physical) model parameter values are uncertain. Often, however, for example in geometrically complex structures with many mechanical connections, it may be far from trivial to identify which physical parameters are causing differences between the numerical and experimental target quantities. In those situations, generic parameters may be better candidates for model updating. Most of the modal-based model updating techniques are concerned with updating the resonance frequencies [2]. Recently, some effort has been dedicated to include the effect of zeros (antiresonances) in the model updating, see [7, 8]. However, all of these model updating techniques are based on updating physical design parameters, while in IPZ model updating, the eigenvalues of the stiffness and/or damping matrix of the (sub)structure are introduced to be the design parameters. This is done because errors in stiffness/damping modeling are more likely to occur than errors in mass modeling. In general, IPZ model updating tries to improve a reduced FE model by minimizing residuals between the numerical and the experimental values of poles and zeros of an FRF. This is done by iteratively updating the eigenvalues of the M. Dorosti () • R.H.B. Fey • H. Nijmeijer Mechanical Engineering Department, Eindhoven University of Technology, 5600 MB Eindhoven, The Netherlands e-mail: [email protected]; [email protected]; [email protected] M.F. Heertjes Mechanical Engineering Department, Eindhoven University of Technology, 5600 MB Eindhoven, The Netherlands ASML, De Run 6501, 5504 DR Veldhoven, The Netherlands e-mail: [email protected] M.M.J. van de Wal ASML, De Run 6501, 5504 DR Veldhoven, The Netherlands e-mail: [email protected] © The Society for Experimental Mechanics, Inc. 2017 J. Caicedo, S. Pakzad (eds.), Dynamics of Civil Structures, Volume 2, Conference Proceedings of the Society for Experimental Mechanics Series, DOI 10.1007/978-3-319-54777-0_8

65

66

M. Dorosti et al.

stiffness/damping matrix of the (sub)structure as generic parameters. In this paper, IPZ model updating is extended to incorporate multiple FRFs. Using multiple FRFs from different actuator/sensor configurations helps to better improve the overall accuracy of the reduced FE model toward the real system since more information on the eigenmodes is used. The remainder of the paper is organized as follows. IPZ model updating using multiple FRFs is discussed in Sect. 8.2. In Sect. 8.3, as a case study, a simulated pinned-sliding beam structure is introduced on which the IPZ model updating technique using multiple FRFs is verified. Finally, some conclusions are drawn in Sect. 8.4.

8.2 IPZ Model Updating Using Multiple FRFs Updating a full FE model is computationally expensive. In addition, the physical parameters causing the difference between the model and the real system may not be known. In this case, updating a reduced-order FE model using generic parameters becomes useful. Using a model reduction technique based on eigenmodes and residual flexibility modes, consider the following reduced-order dynamic equation of an FE model: Mr qR r C Br qP r C Kr qr D fr ;

(8.1)

where qr is a set of desired DOFs including actuators, sensors, and unmeasured performance variables DOFs. Mr ; Br ; Kr 2 Rrr are the reduced order mass, damping, and stiffness matrices and fr 2 Rr1 is the reduced external load column. Using the reduced-order dynamic equation in (8.1), the FRF corresponding to a sensor at DOF i and an actuator at DOF j can be described as Gij .!/ D

det.! 2 Ms C j!Bs C Ks / ; det.! 2 Mr C j!Br C Kr /

(8.2)

where Ms ; Bs ; Ks are the so-called substructure matrices which are constructed from the reduced-order matrices Mr ; Br ; Kr respectively, by eliminating the ith row and the jth column corresponding to the sensor and actuator DOFs. Therefore, Kr can be written as Kr D Kr;s C Kr ;

(8.3)

where Kr;s is partitioned such that 2

Kr;s

3 Ks;1 0 Ks;2 D 4 0 0 0 5; Ks;3 0 Ks;4

(8.4)

and 3 0 kr;1 0 Kr D 4 kr;2 kr;3 kr;4 5 ; 0 kr;5 0 2

(8.5)

Ks is derived from (8.4), i.e. " Ks D

# Ks;1 Ks;2 : Ks;3 Ks;4

(8.6)

In IPZ model updating, a selected number of the eigenvalues of the stiffness matrix Kr and the substructure stiffness matrix Ks are introduced to represent the design parameters  p and  z for updating poles and zeros, respectively. The poles of the system (p ) are the roots of the denominator in (8.2), while the zeros of FRF Gij (z ) are the roots of the numerator.

8 Iterative Pole-Zero Model Updating Using Multiple Frequency Response Functions

67

Now assume that m-FRF measurements from different actuator/sensor configurations are available. In IPZ model updating using multiple FRFs, the quadratic pole-zero error functional 02

3 2 02 3 2 3 1H 31 p;e p;n ./ p;e p;n ./ B6 z ;e 7 6 z ;n ./ 7 C B6 z ;e 7 6 z ;n ./ 7 C B6 1 7 6 1 B6 1 7 6 1 7C 7C

i ./ D B6 : 7  6 7 C W  B6 : 7  6 7 C; : :: : : @4 : 5 4 @4 :: 5 4 5A 5A : : zm ;e

zm ;n ./

zm ;n ./

zm ;e

i

(8.7)

i

is iteratively minimized by updating the generic parameters  T D ΠTp ;  Tz1 ; : : :;  Tzm , where the subscripts e and n indicate experimentally and numerically obtained quantities, respectively. z1 ; : : :; zm stand for the set of zeros from the first up to the mth-FRF measurement, respectively. The diagonal weighting matrix W  0 is applied in order to have equal contributions of the relative errors from each pole and zero. In a linear time invariant (LTI) system, pole locations are identical in any FRF measurement of the system. Therefore, in (8.7), pole residuals is used once, whereas the zero residuals are repeated for each FRF measurement since zero locations vary per actuator/sensor configuration. Replacing the entries p;n ./ and z1 ;n ./ up to zm ;n ./ in (8.7) by their first-order Taylor series approximations around  i results in the approximation of the error functional, or

i ./ D rH i ./W ri ./;

(8.8)

where 3 2 3 2 @p;n ./ p;n ./ p;e @ p 6 z ;e 7 6 z ;n ./ 7 6 0 6 1 7 6 1 7 6 ri ./ D 6 : 7  6 7 6 :: 6 4 :: 5 4 5 : 4 ::: zm ;e zm ;n ./ i 0 „ ƒ‚ … „ 2

i

3 2 3  p 7 6 @z1 ;n ./  z1 7 ::: 0 7 7 @ z1 7 6 :: 7 7 6 4 ::: ::: ::: 5 : 5: @zm ;n ./  zm i 0 : : : @ z m ƒ‚ …i „ ƒ‚ … 0

:::

0

(8.9)

 i

Si

Si in (8.9) represents the sensitivity matrix which includes the sensitivity of each pole denoted by @p;n .p / D @p 2p;n

p T p

T @Kr @p

Mr

p

p

C

T p

;

Br

p

Bs

z

(8.10)

together with the sensitivity of each zero @z;n .z / D @z 2z;n

z T z

T @Ks @z

Ms

z

zC

T z

:

(8.11)

The number of design parameters is generally equal or lower than the number of measured poles and zeros. Therefore, the optimization problem is (over)determined; hence, a least squares approach is applied. Minimizing (8.8) with regard to  i , and requiring that @ i =@ i D 0, leads to the following equation [9]:    H  Re SH i W Si  i D Re Si W i :

(8.12)

The iteration process will be terminated using the following stop criterion: j i  i1 j  ı; where ı is a sufficiently small number. In summary, the proposed IPZ model updating algorithm using multiple FRFs will be given as follows: 1. Preparation (a) Choose the frequency range of interest; identify the experimental values of the poles and zeros in this range.

(8.13)

68

M. Dorosti et al.

(b) Select the desired physical DOFs qr including the actuators, sensors, and unmeasured performance variables DOFs; then apply the model reduction step. (c) Match the experimental poles and zeros with their numerical counterparts from the reduced FE model. (d) Select the stop criterion parameter ı in (8.13). 2. IPZ model updating (a) Select a number of (lowest) eigenvalues of the reduced stiffness matrix Kr and the substructure reduced stiffness matrices Ks1 ; : : :; Ksm , corresponding to the first up to the mth FRF, respectively, as the generic parameters  p ;  z1 ; : : :;  zm . (b) Calculate the sensitivity matrix Si in (8.9) based on (8.10), (8.11) and calculate i in (8.9). (c) Calculate  i according to (8.12), then update the design parameter  i with  i . (d) Incorporate the updated generic parameters  p ,  z1 ; : : :;  zm into Kr ; Ks1 ; : : :; Ksm , respectively, to calculate the updated structure and substructure stiffness matrices Kur and Kus1 ; : : :; Kusm , respectively. (e) Use (8.3)–(8.6) to replace Kus1 into Kur and construct the updated stiffness matrix Kur;s1 ; this process repeats until the replacement of Kusm in Kur;sm1 to construct the final updated stiffness matrix Kur;sm . 3. Stop criterion check (a) Calculate i in (8.7) using the updated stiffness Kur;sm instead of Kr ; check whether the stop criterion in (8.13) is satisfied; if not, return to step 2 (c) using Kur;sm instead of Kr ; the iteration process stops when the stop criterion in (8.13) is satisfied.

8.3 Case Study: 2D Pinned-Sliding Beam To verify the IPZ model updating using multiple FRFs consider a 2D finite element model of a pinned-sliding aluminum beam, see Fig. 8.1, with properties listed in Table 8.1. In the model, twelve Euler beam elements of equal size are used. Each element has two nodes and each node has two DOFs: transversal (w) and rotational (r) displacement. Modal damping of 0:1% is added to all modes. This model is referred to as the original model Gn . Assume that the original model Gn differs from the experimental structure Ge in the 12th element which in reality has 3:0  102 m thickness, and 4:0=3:0  102 m width. Therefore, the original and experimental model have identical mass. A force actuator u is located on the transversal DOF of the third node. Two displacement sensors are located on the transversal DOF of the fourth (y1 ) and the fifth (y2 ) node, while the unmeasured performance variable (z) is represented by the transversal DOF of the seventh node. The following FRFs are introduced: G1 D Y1 =U, G2 D Y2 =U, and G3 D Z=U. In reality, the performance variables cannot be measured directly. However, in this simulated experiment, the quality of the updated performance variable (G3 ) can be assessed since the “experimental” Ge;3 can be calculated. It will be assumed that in a real experiment all the experimental poles and zeros can be estimated from the measured FRFs by applying modal parameter fit procedures. In the simulated experiment, they can simply be calculated. Note that the complex poles/zeros are assumed to occur in complex conjugated pairs since weak damping is assumed. In what follows, only the poles/zeros with positive imaginary parts will be used.

u y1 y2

z

n3

n7

n4 n5

e12

w

r

Fig. 8.1 2D Pinned-Sliding aluminum beam Table 8.1 Properties of the beam

Property Young’s modulus Mass density Thickness Width Length of the beam

Value 69 2:7  103 2:0  102 2:0  102 1:0

Unit GPa kg m3 m m m

G3 [m/N] in dB

G2 [m/N] in dB

G1 [m/N] in dB

8 Iterative Pole-Zero Model Updating Using Multiple Frequency Response Functions

69

0

−100

−200

102

103

102

103

102

103

104

0

−100

−200

Gr Ge Gur

0

104

−100

−200 104

Frequency [Hz] Fig. 8.2 Original, experimental, and updated G1 (top), G2 (middle), and G3 (bottom)

log10 ( i )

8 6 4 2

0

1

2

3

4

Iteration

5

6

7

8 x 105

Fig. 8.3 log10 . i / for IPZ model updating using G1 and G2

It is assumed that Ge;1 and Ge;2 are available and that the frequency range of interest is Œ40; 5000 Hz. Therefore, the original model Gn will be reduced to Gr considering the first five eigenmodes plus one residual flexibility mode for the truncated modes (modes: 6; : : :; 24). A coordinate transformation is applied so that the reduced model is expressed in terms of the desired DOFs qr D Œw4 ; w6 ; w8 ; w12 ; w16 ; w20 ; containing the actuator, sensors, and the unmeasured performance variables DOFs. The goal is to match the poles and the zeros of the reduced-order Gr;1 and Gr;2 with the poles and the zeros of the experimental Ge;1 and Ge;2 , respectively, in the frequency range of interest. According to Fig. 8.2, the first four poles of Ge;1 and Ge;2 , the first three zeros of Ge;1 , and the first two zeros of Ge;2 lay within the frequency range of interest, hence are assumed to be extracted from the measurements. The first four poles of Gr;1 and Gr;2 are updated using the first four (real) eigenvalues of the reduced-order stiffness matrix Kr , i.e. Œp;1 ; : : :; p;4 . Furthermore, the first three zeros of Gr;1 and the first two zeros of Gr;2 are updated using the (real) eigenvalues Œz1 ;2 ; : : :; z1 ;4  and Œz2 ;3 ; z2 ;4  of the substructure reduced-order stiffness matrices Ks1 and Ks2 , respectively. Using the discussed IPZ model updating algorithm with ı D 105 , the algorithm converged to i D 244:58 in i D 755;634 iterations. Although many iterations are needed, the required calculation time is limited because a reduced-order model is used. Note that the number of iterations also depends on the value of the stop criterion parameter ı. It can be seen from Fig. 8.3 that i decreased rapidly during the first iterations, but it takes many iterations to fulfill the convergence criterion of (8.13) with ı D 105 . Convergence of the IPZ model updating toward a (local) minima of (8.7), results in the updated reduced model Gu composed of Mr , Br , and Kur;s2 . Figure 8.2 shows that Gu;1 and Gu;2 match very well in terms of both poles and zeros with the poles and the zeros of Ge;1 and Ge;2 in the frequency range of interest. Moreover, Gu;3 which resembles the unmeasured performance variable also matches very well in terms of both the poles and the zeros with those from Ge;3 .

70

M. Dorosti et al.

Table 8.2 Correlations in the frequency range of interest before/after IPZ model updating using G1 and G2 XS XA

Gn;1 vs. Ge;1 0:025 0:004

Table 8.3 Correlations in the frequency range of interest before and after IPZ model updating using only G1

Gur;1 vs. Ge;1 0:999 0:994

Gn;2 vs. Ge;2 0:025 0:004

Gur;2 vs. Ge;2 0:991 0:993

Gn;3 vs. Ge;3 0:026 0:004

Gur;3 vs. Ge;3 0:669 0:994

Gn;1 vs. Ge;1 Gu;1 vs. Ge;1 Gn;3 vs. Ge;3 Gu;3 vs. Ge;3 XS 0:025 0:946 0:026 0:336 XA 0:004 0:950 0:004 0:954

Furthermore, the shape (XS ) and amplitude (XA ) correlation measures [6], listed in Table 8.2, show significant improvement after IPZ model updating not only for the FRFs which are used in the model updating (G1 and G2 ), but also for the unmeasured FRF (G3 ). Correlation measures for the case where only G1 is included in the model updating process is listed in Table 8.3. The results indicate that correlations for both G1 and G3 are higher when an extra FRF G2 is used in the model updating process. This means the reduced FE model is indeed better improved, when more FRFs are included the model updating process.

8.4 Conclusion In this paper, the IPZ model updating technique is extended to enable the use of multiple FRFs so that more information of eigenmodes can be used. The IPZ model updating using multiple FRFs is verified based on a simulated experiment of a pinned-sliding beam. It was shown that using more FRF measurements, from different actuator/sensor configurations, helps to better improve the reduced FE model toward the (simulated) real system. After IPZ model updating, the numerical FRFs match very well with the experimental FRFs. This not only holds for the two experimental FRFs which were used in the model updating procedure, but also for the third FRF related to an unmeasured performance variable. Although the method seems promising, there are still unresolved issues in the proposed model updating method which need further investigation: (1) a proof of convergence to a local minimum, although the convergence behavior shown in Fig. 8.3 is promising, a rigorous proof is lacking and (2) after model updating the stiffness matrix shows a small amount of asymmetry.

References 1. Yang, J., Ouyang, H., Zhang, J.-F.: A new method of updating mass and stiffness matrices simultaneously with no spillover. J. Vib. Control 22(5), 1181–1189 (2016) 2. Mottershead, J.E., Link, M., Friswell, M.I.: The sensitivity method in finite element model updating: a tutorial. Mech. Syst. Signal Process. 25, 2275–2296 (2011) 3. Jaishi, B., Ren, W.-X.: Finite element model updating based on eigenvalue and strain energy residuals using multiobjective optimisation technique. Mech. Syst. Signal Process. 21, 2295–2317 (2007) 4. Dorosti, M., Fey, R., Heertjes, M., van de Wal, M., Nijmeijer, H.: Finite element model reduction and model updating of structures for control. In: 19th World Congress The International Federation of Automatic Control, Cape Town (2014) 5. Arora, V., Adhikari, S., Friswell, M.: FRF-based finite element model updating method for non-viscous and non-proportional damped system. In: International Conference on Structural Engineering Dynamics, Lagos (2015) 6. Dorosti, M., Heck, F., Fey, R., Heertjes, M., van de Wal, M., Nijmeijer, H.: Frequency response sensitivity model updating using generic parameters. In: American Control Conference (ACC) (2016) 7. Arora, V.: Constrained antiresonance frequencies based model updating method for better matching of FRFs. Inverse Prob. Sci. Eng. 22(6), 873–888 (2014) 8. D’Ambrogio, W., Fregolent, A.: Results obtained by minimising natural frequency and antiresonance errors of a beam model. Mech. Syst. Signal Process. 17(1), 29–37 (2003) 9. de Kraker, B.: A Numerical-Experimental Approach in Structural Dynamics. Shaker, Maastricht (2013)

Chapter 9

Vision-Based Concrete Crack Detection Using a Convolutional Neural Network Young-Jin Cha and Wooram Choi

Abstract The prominent methods for monitoring structures to date rely on analyzing data measured from contact sensors that are physically attached to a structure. However, these approaches have the high possibility of false alarms due to noises, sensor malfunctions, and complex environmental effects. Under those circumstances, engineers have to conduct on-site investigations to confirm that damage has occurred. To address this challenge, this paper proposes a new visionbased approach for detecting concrete cracks using a convolutional neural network (CNN). Images are firstly taken under uncontrolled situations to collect widely varying crack features. Second, the raw images are divided into 40K images to build training and validation sets. Lastly, the prepared datasets are fed into a deep CNN architecture with eight layers including convolution, pooling, ReLU, and softmax. The trained classifier consequently records 98% of accuracies in both training and validation. Keywords Concrete crack detection • Convolutional neural network • Structural health monitoring • Vision-based • Machine learning

9.1 Introduction The deterioration of civil structures is an inevitable process occurring from the beginning stage of the use. Considering a single damaged member can cause significant changes [1] in their systems, structural health monitoring (SHM) cannot be overvalued. Therefore, structural safety has become a major issue, and routine inspections are regularized [2] as a result. However, heavy reliance on human-conducted on-site inspection arises the following problems. First, the considerable number of structures hinders frequent inspections due to limited human-resources [3]. Second, evaluation results by humaninspectors differ dependent on their physical ability, knowledge, and experience [3, 4]. Hence, developing automated and reliable inspection methods is necessary. In order to overcome the above issues, contact-sensor approaches have been developed [5, 6]. One of the particular works focused on monitoring the global behavior of a suspension bridge was conducted by Kurata et al. [7]. Wireless sensors, including tri-axial accelerometers, potentiometers, and climate stations, collected data from each sensor node to extract seven mode shapes of the bridge. Although these research articles are highlighted as accurate measurements in the field, the complex requirements, such as installing multiple sensors, integration of each data and maintenance of sensors, remains as challenges. Obtaining adequate data is also known as one of the toughest issues due to environmental variations [8, 9] especially temperature and humidity. This aspect may require extra efforts to identify modal properties accurately [10]. Vision-based approaches have emerged as the alternatives because the results provide intuitively comprehensive information compared to sensor-based methods. Several modal identification methods using high-speed cameras have been proposed [11, 12]. However, further tests for massive structures are required to completely substitute sensor-based monitoring systems. While the main purpose of the aforementioned studies is lying on observing the global behaviors of structures, several research projects that mainly dedicate to detecting local damages from images have been proposed [13, 14]. Although the vision-based damage detection partly overcome some limitations of sensor-based approaches, it is still challenging to create methods that actually works in real-world situations.

Y.-J. Cha () Department of Civil Engineering, University of Manitoba, Winnipeg, MB, R3T 5V6, Canada e-mail: [email protected] W. Choi Department of Civil Engineering, University of Manitoba, Winnipeg, MB, R3T 6B3, Canada e-mail: [email protected] © The Society for Experimental Mechanics, Inc. 2017 J. Caicedo, S. Pakzad (eds.), Dynamics of Civil Structures, Volume 2, Conference Proceedings of the Society for Experimental Mechanics Series, DOI 10.1007/978-3-319-54777-0_9

71

72

Y.-J. Cha and W. Choi

Fig. 9.1 Overall architecture

A possible breakthrough is the implementation of artificial neural networks [15]. Especially, convolutional neural networks (CNNs) have received a great attention in image classification due to the excellent performance [16]. In addition, extremely fast computations by the conjugation of graphic process units (GPUs) enable CNNs to have deeper architectures that can learn a vast amount of features from datasets [17]. In this study, a deep CNN architecture is proposed for detecting concrete cracks, and two GPUs (Nvidia Geforce Titan X  2ea) are used [18].

9.2 Overall Architecture of the Proposed CNN and the Results The designed architecture is depicted in Fig. 9.1 representing how an input data is generalized by passing through the designed architecture. Overall, input images with 256  256  3 pixel resolutions are generalized to the vector with 96 elements in training, and the vector is consequently classified to crack or intact after rectified linear unit (ReLU), the last convolution, and softmax layers. During the training process, pooling layers simply pool features from previous layers. These operations may reduce computational cost significantly. Convolution layers conduct element by element multiplications between inputs of each layer and receptive fields, where the values of receptive fields are given by random numbers and tuned in stochastic gradient descent. The proposed CNN trained on 32K images and 8K extra images are used in validation, in which the images were cropped from raw images taken under uncontrolled conditions to generate a dataset with extensively varying features of cracks. As a result, the designed architecture records 98.25 and 98.04% of accuracies in training and validation respectfully. It is anticipated that the trained CNN classifier works very well in testing with images that are taken under uncontrolled situations because the generated training dataset includes extensively varying crack images, and the accuracies reach to about 98%. In the future, the trained CNN will be combined with a sliding window technique and autonomously flying drones.

9.3 Conclusion A vision-based method for detecting concrete cracks using a deep CNN was proposed. To generate the dataset with extensively varying crack features, images were taken under uncontrolled circumstances, and they cropped into small images of 256  256. The generated image database was fed into the designed CNN. Consequently, 98% of accuracies in training and validation were recorded. In the future, we will combine the trained CNN with a sliding window technique and autonomously flying drones to develop an advanced framework.

9 Vision-Based Concrete Crack Detection Using a Convolutional Neural Network

73

References 1. Begg, R.D., et al.: Structural integrity monitoring using digital processing of vibration siqnals. In: Offshore Technology Conference: Offshore Technology Conference (1976) 2. Administration, F.H. [cited 2016 September 21]; Available from: https://www.fhwa.dot.gov/bridge/ 3. Chang, P.C., Flatau, A., Liu, S.: Review paper: health monitoring of civil infrastructure. Struct. Health Monit. 2(3), 257–267 (2003) 4. Phares, B.M., et al.: Reliability of visual bridge inspection. Public Roads. 64(5), (2001) 5. Rice, J.A., et al.: Flexible smart sensor framework for autonomous structural health monitoring. Smart Struct. Syst. 6(5–6), 423–438 (2010) 6. Jang, S., et al.: Structural health monitoring of a cable-stayed bridge using smart sensor technology: deployment and evaluation. Smart Struct. Syst. 6(5–6), 439–459 (2010) 7. Kurata, M., et al.: Internet-enabled wireless structural monitoring systems: development and permanent deployment at the New Carquinez Suspension Bridge. J. Struct. Eng. 139(10), 1688–1702 (2012) 8. Xia, Y., et al.: Temperature effect on vibration properties of civil structures: a literature review and case studies. J. Civ. Struct. Heal. Monit. 2(1), 29–46 (2012) 9. Cornwell, P., et al.: Environmental variability of modal properties. Exp. Tech. 23(6), 45–48 (1999) 10. Behmanesh, I., et al.: Hierarchical Bayesian model updating for structural identification. Mech. Syst. Signal Process. 64, 360–376 (2015) 11. Dorn, C.J., et al.: Automated extraction of mode shapes using motion magnified video and blind source separation. In: Topics in Modal Analysis & Testing. Springer, Cham/Heidelberg (2016) 12. Chen, J.G., et al.: Modal identification of simple structures with high-speed video using motion magnification. J. Sound Vib. 345, 58–71 (2015) 13. Abdel-Qader, I., Abudayyeh, O., Kelly, M.E.: Analysis of edge-detection techniques for crack identification in bridges. J. Comput. Civ. Eng. 17(4), 255–263 (2003) 14. Cha, Y.-J., You, K., Choi, W.: Vision-based detection of loosened bolts using the Hough transform and support vector machines. Autom. Constr. 71, 181–188 (2016) 15. LeCun, Y., et al.: Gradient-based learning applied to document recognition. Proc. IEEE. 86(11), 2278–2324 (1998) 16. LeCun, Y., Bengio, Y., Hinton, G.: Deep learning. Nature. 521(7553), 436–444 (2015) 17. Steinkrau, D., Simard, P.Y., Buck, I.: Using GPUs for machine learning algorithms. In: 8th International Conference on Document Analysis and Recognition. IEEE, Seoul, Korea (2005) 18. Cha, Y.-J., Choi, W.: Deep learning-based crack detection using convolutional neural networks. Comput. Aided Civ. Infrastruct. Eng. 32(3) (2017). doi:10.1111/mice.12263

Chapter 10

Analytical and Experimental Analysis of Rocking Columns Subject to Seismic Excitation Ryan Kent Giles and Thomas John Kennedy

Abstract Structures of ancient Mediterranean cultures that have survived numerous earthquakes over the span of millennia relied on multifaceted rocking columns to dissipate seismic energy. Rocking columns are again emerging as an effective mechanism in modern structural systems; understanding the rocking behavior can provide insight into how to best design this type of system. This study examines the analytical and experimental rocking behavior of a rectangular column. Equations of motion that describe the rocking behavior of the polygonal columns are derived and analytical energy dissipation methods compared. A high-speed 3D motion capture system, providing noncontact measurement of the column motion, is used in a series of experiments on a uniaxial shake table to validate the analytical model. These experiments show variation indicating stochastic behavior during the excitation phase. The damping ratio and coefficient of restitution are calculated from the experimental results. The experimental results and analytical solution are compared. Keywords Rocking columns • Coefficient of restitution • Impact damping • 3D metrology • Seismic excitation

10.1 Introduction The ability of ancient Greek temples and other structures to survive seismic events is due to the rocking motion of their columns. When stripped of their architectural adornments and cultural significance, columns are slender, rigid blocks. Many researchers have derived the equations of motion for slender, rigid blocks in two dimensions [1–8] and performed analytical analysis of their behavior under various sources of base excitation [9–13]. The effects of friction and shape of the initiation of rocking or sliding have also been well studied analytically [14]. There has been some extension of the rocking problem to three dimensions accounting for cylinders [4] and general three-dimensional structures with a rectangular base [15, 16]. Fewer studies have experimentally confirmed these analytical equations [2]. This paper presents an experimental validation of the equations of motion of a slender rigid body under seismic excitation. The results draw attention to the typical analytical assumptions for the coefficient of restitution and the probabilistic nature of experiments.

10.2 Background For the case of two-dimensional rocking (see Fig. 10.1), the block is assumed to be located on a rigid base. The column has a height of 2h and a width of 2b. We define a slender column as one where: h 1 >p b 1 C IO

(10.1)

given IO is the moment of inertia of the block about O or O0 which for a rectangle is equal to (4/3)mR2 . The slenderness of the block controls whether, on impact, the rigid block will switch its rotation point from O or O0 or bounce back and continue to rotate about O. The behavior of interest for structural columns is the switching of the rotation point upon impact.

R.K. Giles () • T.J. Kennedy Department of Civil Engineering, Stony Brook University, 2426 Computer Science, Stony Brook, NY, 11794-4424, USA e-mail: [email protected] © The Society for Experimental Mechanics, Inc. 2017 J. Caicedo, S. Pakzad (eds.), Dynamics of Civil Structures, Volume 2, Conference Proceedings of the Society for Experimental Mechanics Series, DOI 10.1007/978-3-319-54777-0_10

75

76

R.K. Giles and T.J. Kennedy

Fig. 10.1 Rigid rectangular block freely rocking on a rigid base

The derivations also assume that the coefficient of friction between the column and the rigid base is sufficient to prevent sliding. Therefore, the horizontal ground motion (üg ) may cause either rigid motion with the column or rocking of the column. If rocking initiates, as in Fig. 10.1, the block will move about the center of rotation (O) with an angle of rotation (). The equations of motion [1, 2, 10, 12] on the block are: IO R C mgRsin .˛  / D mRcos .˛  / uR g

(10.2)

when the block is in rotation about O. When it is rocking about the opposite corner, (O0 ), then the equation of motion becomes: IO R C mgRsin .˛   / D mRcos .˛  / uR g

(10.3)

p In both equations, ˛ is equal to tan1 (b/h) and R is equal to (b2 C h2 ). Both of these equations can be combined taking advantage of the signum function and solving for the angular acceleration.   uR g R D p2 sin .˛sgn Œ  / C cos .˛sgn Œ    / g

(10.4)

p where p is (3g/4R) and a dynamic characteristic related only to the block’s geometry. When the column impacts with the base (i.e.  D 0), there must be a conservation of momentum but there is also an opportunity for energy loss in the system. Yim et al. [1] derived a coefficient of restitution, e, such that: PC. e D  P 

(10.5)

where P  and P C are the angular velocity just before and after impact. The properties of the rectangular column are such that the coefficient of restitution can be analytically calculated to be 3 e D 1  sin2 ˛ 2

(10.6)

The energy loss due to impact is therefore 1e2 and the higher the coefficient of restitution, the smaller the energy loss and the less damping takes place in the system. Equation (10.6), as written depends solely on the geometry of the column. More recently, Calio and Marletta [6] gave an alternative expression for the coefficient of restitution:

10 Analytical and Experimental Analysis of Rocking Columns Subject to Seismic Excitation

. eDh R

77

(10.7)

Again, Eq. (10.7) is dependent solely on the geometry. However, Aslam et al. [2], in their experimental results, purposely did not formally derive or give any expression for the coefficient of restitution. Instead, they stated that the coefficient of restitution would depend on P  and the material properties of the column. This research considers the coefficient of restitution to be a primary parameter for matching the experimental and analytic results.

10.3 Analytical Model The equation of motion for the rocking column (Eq. (10.4)) was programmed into MATLAB using its explicit Runge-Kutta method based on the Dormand-Price (4,5) pair. To model the impact conditions, the event finder in the method was used such that when  D 0, the solver would stop, reduce the angular velocity by the coefficient of restitution, and restart the solver with the updated initial conditions ( D 0 and P C D eP  ). The coefficient of restitution was not programmed explicitly to be either Eq. (10.6) or Eq. (10.7). Instead, it was left as a parameter that could be changed to allow the model to be updated according to the experimental results.

10.4 Experimental Method The column used in this study was a 2.50 cm  2.50 cm  25.4 cm (0.985 in  0.985 in  10.0 in) machined steel rod weighing 1.238 kg (2.729 lbs). The column was painted with a series of tracking points on one side at regular 6 cm intervals starting 4 cm from the bottom of the column, as seen in Fig. 10.2. One end was marked to indicate the top of the column. A uniaxial Quanser Shake Table II provided the horizontal base excitation. A sheet of 400 grit sand paper was mounted on the shake table to increase the coefficient of friction between the shake table surface and the column. The increased coefficient of friction prevented sliding and rocking-sliding motion. The column was placed at the center of the shake table and aligned so that the axis of the shake table was perpendicular to the sides of the rectangular column using a carpenter’s square. The excitation commanded by the shake table was a modified East-West strong motion record for the 1986 Kalamata earthquake. The record was scaled to initiate rocking in the column but not topple the column. The record was clipped as the first part of the record provides essentially one large pulse and one smaller pulse in the displacement of the table as seen in Fig. 10.3a. Artificial pulses have been previously used in analytical evaluations of rocking columns [13]. Figure 10.3b shows the corresponding accelerations the command displacement produced. Fig. 10.2 Experimental setup showing steel column and the tracking results of four points on the column and one on the shake table

78

R.K. Giles and T.J. Kennedy

a

b 0.1

0 Acceleration (g)

Command Displacement (cm)

0.2

−0.2 −0.4

0

−0.05

−0.6 −0.8

0.05

0

0.5

1

1.5 Time (s)

2

2.5

3

−0.1

0

0.5

1

1.5

2

2.5

3

Time (s)

Fig. 10.3 (a) Command displacement (cm) for the Quanser Shake Table II. (b) Corresponding command accelerations (g)

Noncontact, three-dimensional measurement of the column motion was performed using a three-camera Xcitex ProCapture system. The cameras were arranged to capture the three-dimensional movement of the column. Each camera operated at a sampling speed of 125 frames per second. Recordings were individually calibrated within Xcitex ProAnalyst software to remove lens distortion and merged in pairs to allow for the two-dimensional images to be interpreted as threedimensional measurements. The ProAnalyst software is able to track each point individually over the entire length of the record. Four points on each column and a fixed point on the shake table were tracked for each trial. Figure 10.2 shows the four column points and the table point tracked and overlaid on the image of the column at rest. An accelerometer on the shake table was synchronized to the image data to confirm the desired command accelerations. Positional data was exported from the ProAnalyst software and displacement measurements of the column relative to the shake table surface were calculated. The data was also filtered using an eight pole elliptical filter with a 50 Hz cutoff frequency. The filter served to reduce the small amount of jitter inherently present due to the motion tracking. Anti-alias filtering is not possible with optical systems and therefore oversampling above the expected behavior of the rocking is required to produce accurate results.

10.5 Results and Discussion Thirty-two experiments were performed and processed for this analysis. As the cameras are located in fixed positions external to the shake table system, the motion of the table was removed from the motion of the column to generate displacement values relative to the shake table. Figures 10.4, 10.5, 10.6 and 10.7 show typical results for the tracking of the top tracking point (i.e. Point 1 in Fig. 10.2) in all three directions—x is the axis of the table motion, y is the gravitational axis, and z is out of plane. These figures have been plotted in the same scale to emphasize the differences between the records. The tracked motion of the table is shown in each plot for reference. The rocking motion in the x-axis is greater than the other directions as expected from the desired experimental focus on rocking in one direction. To investigate the differences apparent in the experiments, the free rocking of each record was examined further. The rocking records were processed to determine when the forced motion of the table ended and free rocking began. The value for the angular displacement and angular velocity were determined for the initiation of the free rocking. These values served as initial conditions for comparison to the analytical free rocking model. The values for the angular displacement and angular velocity at the end of the table motion showed great variety. Each experiment was excited using the same command displacements as confirmed by the tracking of the table during each trial. Nevertheless, the excitation had varied effects on the forced motion of the column in each trial. The differences are due to slight, uncontrollable, experimental conditions. However, the free rocking of the columns shows similar behavior once it begins. After determining the start of free rocking, the peaks of the motion were located. The initial six peaks were used to calculate the damping ratio using the logarithmic decrement method. A histogram of the damping ratios is presented in Fig. 10.8. Figure 10.9 plots the calculated damping ratio against a lognormal distribution. Both Figs. 10.8 and 10.9 indicate that four of the experiments do not fall within the same distribution of damping coefficients. These four do however appear similar to one another and Fig. 10.6

10 Analytical and Experimental Analysis of Rocking Columns Subject to Seismic Excitation

79

20 Table X Axis Y Axis Z Axis

15

Displacement (mm)

10 5 0 −5 −10 −15 −20

0

2

4

6

8 Time (s)

10

12

14

Fig. 10.4 Representative sample data characterized by a long period of free rocking 20 Table X Axis Y Axis Z Axis

15

Displacement (mm)

10 5 0 −5 −10 −15 −20

0

2

4

6

8 Time (s)

10

12

14

Fig. 10.5 Representative data characterized by a medium length free rocking period

represents this group. They are characterized by a rocking motion that is arrested by the last movement of the table causing a very small initial angle and velocity at the start of free rocking phase. The rocking is quickly damped out by the impacts. The free rocking phases of the experimental results were compared to the results from the analytical model. For the steel column used in the experiments, Eq. (10.6) would yield a coefficient of restitution of 0.9857; Eq. (10.7) would yield a coefficient of restitution of 0.9952; and Aslam et al. [2] found a coefficient of restitution of 0.9250 for their concrete block. Because the damping ratio varied among the samples, so did the coefficient of restitution that best matched the analytical and experimental models. For the data presented in Fig. 10.5, a coefficient of restitution of 0.9750 produced the best result and the comparison in presented in Fig. 10.10. At the start of this plot, there is excellent correlation between the experimental and analytical results as the orange and blue lines lay on top of each other. However, as the column continues to rock, the frequency of its rocks increases as the amplitude of the rock decreases suddenly. At this point, the analytical and experimental results diverge. The analytical results, with a constant coefficient of restitution, continues to dampen at a steady rate. This type of result is typical for all the comparisons between the analytical and experimental results. It is indicative that the coefficient of restitution for a rocking column is a function of the rocking period and the angular velocity.

80

R.K. Giles and T.J. Kennedy 20 Table X Axis Y Axis Z Axis

15

Displacement (mm)

10 5 0 −5 −10 −15 −20

0

2

4

6

8 Time (s)

10

12

14

Fig. 10.6 Representative data characterized by a short free rocking period starting with a small amplitude

20 Table X Axis Y Axis Z Axis

15

Displacement (mm)

10 5 0 −5 −10 −15 −20

0

2

4

6

8

10

12

14

Time (s)

Fig. 10.7 Representative data showing a very large displacement in the forced phase followed by an extended period of free rocking

Fig. 10.8 Histogram of the calculated damping ratios during free rocking of the columns Trials

15

10

5

0

0.005

0.01 0.015 Damping Coefficient

0.02

10 Analytical and Experimental Analysis of Rocking Columns Subject to Seismic Excitation

81

0.995 0.99

Probability

0.95 0.9 0.75 0.5 0.25 0.1 0.05 0.01 0.005 10−3

10−2 Damping Coefficient

Fig. 10.9 Lognormal probability plot of the calculated damping ratios during the free rocking of the columns

Angular Dispalcement, (rad)

1

Experimental Analytical

0.5

0

−0.5

−1 5

5. 5

6

6. 5 7 Time, (s)

7. 5

8

8. 5

Fig. 10.10 Comparison between the experimental and analytical results with an optimized coefficient of restitution

10.6 Conclusion The invention of motion tracking software has made the experimental verification of the analytical solutions of the rocking column possible. The speed and accuracy at which images can be processed has dramatically increased since Aslam et al. performed a few motion capture experiments using hand digitization. The data collected in this study confirmed the conclusions that the coefficient of restitution is not constant, is likely dependent on the angular velocity, and is significantly less than the theoretical derivation. The coefficient of restitution calculated for a steel column in this study is greater than that calculated by Aslam for their concrete blocks. The elastic properties of the two materials, and the geometry of the two columns, can account for the primary differences. Future research will continue to investigate the coefficient of restitution by further classifying the collected data by the relevant parameters at the start of the free rocking. More data will be collected to provide a larger statistical sample that can better accommodate the variation inherent to the experimental process. The analytical solution will be modified to account for the stochastic nature of the excitation phase. The modified analytical model, and its range of results, will be compared to all the collected data.

82

R.K. Giles and T.J. Kennedy

References 1. Yim, C.-S., Chopra, A.K., Penzien, J.: Rocking response of rigid blocks to earthquakes. Earthq. Eng. Struct. Dyn. 8(6), 565–587 (1980) 2. Aslam, M., Scalise, D.T., Godden, W.G.: Earthquake rocking response of rigid bodies. J. Struct. Div. 106(2), 377–392 (1980) 3. Psycharis, I.N., Jennings, P.C.: Rocking of slender rigid bodies allowed to uplift. Earthq. Eng. Struct. Dyn. 11(1), 57–76 (1983) 4. Koh, A., Mustafa, G.: Free rocking of cylindrical structures. J. Eng. Mech. 116(1), 35–54 (1990) 5. Andreaus, U., Casini, P.: On the rocking-uplifting motion of a rigid block in free and forced motion: influence of sliding and bouncing. Acta Mech. 138(3–4), 219–241 (1999) 6. Caliò, I., Marletta, M.: Passive control of the seismic rocking response of art objects. Eng. Struct. 25(8), 1009–1018 (2003) 7. Apostolou, M., Gazetas, G., Garini, E.: Seismic response of slender rigid structures with foundation uplifting. Soil Dyn. Earthq. Eng. 27(7), 642–654 (2007) 8. Contento, A., Di Egidio, A.: Investigations into the benefits of base isolation for non-symmetric rigid blocks. Earthq. Eng. Struct. Dyn. 38(7), 849–866 (2009) 9. Spanos, P., Koh, A.: Rocking of rigid blocks due to harmonic shaking. J. Eng. Mech. 110(11), 1627–1642 (1984) 10. Hogan, S.J.: On the dynamics of rigid-block motion under harmonic forcing. Proc. R. Soc. Lond. A. 425(1869), 441–476 (1989) 11. Hogan, S.J.: The many steady state responses of a rigid block under harmonic forcing. Earthq. Eng. Struct. Dyn. 19(7), 1057–1071 (1990) 12. Makris, N., Roussos, Y.S.: Rocking response of rigid blocks under near-source ground motions. Géotechnique. 50(3), 243–262 (2000) 13. Zhang, J., Makris, N.: Rocking response of free-standing blocks under cycloidal pulses. J. Eng. Mech. 127(5), 473–483 (2001) 14. Shenton III, H.: Criteria for initiation of slide, rock, and slide-rock rigid-body modes. J. Eng. Mech. 122(7), 690–693 (1996) 15. Egidio, A.D., Zulli, D., Contento, A.: Comparison between the seismic response of 2D and 3D models of rigid blocks. Earthq. Eng. Eng. Vib. 13(1), 151–162 (2014) 16. Zulli, D., Contento, A., Di Egidio, A.: 3D model of rigid block with a rectangular base subject to pulse-type excitation. Int. J. Non Linear Mech. 47(6), 679–687 (2012)

Chapter 11

Extending the Fixed-Points Technique for Optimum Design of Rotational Inertial Tuned Mass Dampers Abdollah Javidialesaadi and Nicholas Wierschem

Abstract The fixed-points technique is an approximate version of H-infinity optimization and one of the most common methods used for the design of tuned vibration absorbers. The fixed-points technique is based on the existence of fixed points on the system’s frequency response curve that are independent of the system’s damping level and are thus at the same location during both the zero and infinite damping conditions. Optimum tuning parameter of tuned mass dampers (TMDs) have been obtained by equalizing the magnitude of the response at these fixed-points. This technique has been previously investigated for the optimal design of conventional TMDs and extended for various types of other TMDs. Recently, by replacing the damper in the TMD with combination of a tuning spring, viscous damper, and a small physical mass connected to a mechanism which converts translational motion to the rotational motion of that small mass, rotational tuned mass dampers (RITMDs) have been developed. However, the fixed-points technique has not been extended previously for RITMDs, which have one additional degree-of-freedom compared to TMDs. In this paper, the fixed-points technique is extended, via algebraic solution, for selecting the optimum tuning and damping values of RITMDs. Comparison of the response of the system with the optimum design values determined from the proposed method and numerical results in the literature demonstrates the validity of the assumptions and procedures of the proposed optimization method. Additionally, the performance of the system, in comparison to a conventionally optimized TMD via the fixed-points technique, shows that the RITMDs can be more effective at reducing the underlying system’s maximum displacement response. Keywords Vibration absorber • Tuned mass damper • Fixed-points method • Optimum design • Rotational inertia damper

11.1 Introduction Fixed-points techniques have been proposed and utilized for finding the optimum tuning frequency and damping ratio of tuned mass dampers (TMDs) [1] (Fig. 11.1) as an approximate version of H-infinity optimization. In addition, the fixedpoints method has been used for optimum design of damped non-traditional TMDs [2], developed for TMDs attached to multi-degree of freedom primary systems [3], and generalized for global vibration control [4]. Utilizing the inerter [5], which produces inertial mass through transferring the translational motion of the primary structure to the rotational motion of the inerter, different configurations of rotational mass dampers have been proposed and developed to control the response of the primary structure they are attached to and reduce the physical mass of the absorber [6–9]. As inerter based devices, rotational inertia tuned mass dampers (RITMD) (Fig. 11.2), which consist of a TMD modified with the addition of a tuned of rotational inertia device (RID) such as rack and pinion or ball and screw mechanism, have been introduced and optimized numerically [10]. However, the fixed-point technique has not been developed for optimum design of RITMD. In this paper, a fixed-points technique is developed for optimum design of a RTIMD attached to an undamped SDOF primary structure.

A. Javidialesaadi • N. Wierschem () Department of Civil and Environmental Engineering, The University of Tennessee, Knoxville, TN, USA e-mail: [email protected]; [email protected] © The Society for Experimental Mechanics, Inc. 2017 J. Caicedo, S. Pakzad (eds.), Dynamics of Civil Structures, Volume 2, Conference Proceedings of the Society for Experimental Mechanics Series, DOI 10.1007/978-3-319-54777-0_11

83

84

A. Javidialesaadi and N. Wierschem

ks

k1

ms

m1

c1

cs

x1

xs xo Fig. 11.1 TMD

ks

k1

ms

m1

cs

RID k2 xr

xs

x1

Dynmaic Magnification Factor (DMF)

Fig. 11.2 RITMD 12

Tuned Damping Infinity Damping Zero Damping De-tuned Damping 2 De-tuned Damping 4 De-tuned Damping 3 De-tuned Damping 1

10 8 6 P

Q

4 2 0

0.6

0.8

1.4 1.2 1 Frequency Ratio

1.6

1.8

Fig. 11.3 Fixed-points in the TMD’s frequency response curves

11.2 Fixed-Point Technique for RITMD For traditional TMDs, there are always two fixed-points (P, Q) in the frequency response curve for case of damping equal to zero or infinity [1] (Fig. 11.3). In other words, these points are independent from the damping level of the absorber (c), therefore, the primary structure frequency response curves always passed through these two points. The fixed-points can be found by setting the zero and infinity damping transfer functions equal to each other (kHkc D 0 D kHkc D 1 ). The optimum frequency ratio can be obtained by setting the magnitude of the two fixed points equal. This provides an optimization condition in which the curve has a peak at these points [1]. By putting the partial derivative of frequency response function respect to damping equal to zero at optimum frequency points, two optimum damping are achieved and the final optimum damping is the average of two damping values. Since the RITMD is a three degree-of-freedom system with two optimum frequency ratios, four fixed-points (P1 , P2 , Q1 , Q2 ) in the frequency response curves of the primary structure are observed (Fig. 11.4). This observation can be support mathematically by solving the equation of zero-infinity damping in the RITMD (kHkc D 0 D kHkc D 1 ),

Dynamic Magnification Factor (DMF)

11 Extending the Fixed-Points Technique for Optimum Design of Rotational Inertial Tuned Mass Dampers 12

Zero Damping Infinity Damping Tuned Damping De-tuned Damping 1 De-tuned Damping 2 De-tuned Damping 3 De-tuned Damping 4

10 8 6 4

85

P2

P1

Q1 2 Q2 0 0.6

0.8

1.4 1.2 1 Frequency Ratio

1.6

1.8

Dynamics Magnification Factor (DMF)

Fig. 11.4 Fixed-points in the RITMD’s frequency response curves

4

3

2

1

0 0.5

Proposed Fixed Points Method Nonlinear Programming Optimization

1 Frequency Ratio

1.5

Fig. 11.5 Extended Fixed-Point Method vs Numerical Optimization; m1 D 10% ms (rotational mass D 15% m1)

which is a forth degree polynomial with four real positive roots. Extending the fixed-points method, we assumed the optimum frequencies condition occurs in the case of a pair equality of the primary structure response magnitude in the fixedpoints (kHP1 k D kHQ1 k; kHp2 k D kHQ2 k). This assumption leads to two high-order nonlinear equations which are solved numerically to find the optimum frequency ratios. In the final step, the optimum damping can be found by putting the maximum response magnitude of one of two frequency ratios equal to response magnitude of the other optimum frequency ratio.

11.3 Results and Discussion To examine the proposed extended fixed-points technique accuracy, the frequency response of the primary system with optimum design values from the proposed method is compare with the optimum H-infinity design utilizing an numerical nonlinear programming optimization method [10] (Fig. 11.5). The propose method is an approximate method, thus there are small difference in comparison to the exact H-infinity optimization utilizing nonlinear programming; however, the results from the proposed method are close, which demonstrates the accuracy of the assumptions in the extended proposed method. In addition, the response of the primary structure of both RITMD and TMD systems with the same secondary mass ratio (10%) and optimized with the fixed point method is shown in Fig. 11.6. It can be observe that the RITMD exhibits superior performance in reducing the vibration amplitude of the primary system in compare to the TMD.

A. Javidialesaadi and N. Wierschem

Dynamic Magnification Factor (DMF)

86 5 TMD RITMD

4 3

2 1 0 0.4

0.6

0.8

1 1.2 Frequency Ratio

1.4

1.6

Fig. 11.6 TMD vs RITMD Response; m1 D 10% ms (rotational mass D 15% m1)

References 1. Den Hartog, J.: Mechanical Vibrations, 4th edn. McGraw-Hill, New York (1956) 2. Liu, K., Liu, J.: The damped dynamic vibration absorbers: revisited and new result. J. Sound Vib. 284(3–5), 1181–1189 (2005) 3. Ozer, M.B., Royston, T.J.: Extending Den Hartog’s vibration absorber technique to multi-degree-of-freedom systems. J. Vib. Acoust. 127(4), 341 (2005) 4. Dayou, J.: Fixed-points theory for global vibration control using vibration neutralizer. J. Sound Vib. 292(3–5), 765–776 (2006) 5. Smith, M.C.: Synthesis of mechanical networks: the inerter. IEEE Trans. Autom. Control. 47(10), 1648–1662 (2002) 6. Hwang, J.-S., Kim, J., Kim, Y.-M.: Rotational inertia dampers with toggle bracing for vibration control of a building structure. Eng. Struct. 29(6), 1201–1208 (2007) 7. Ikago, K., Saito, K., Inoue, N.: Seismic control of single-degree-of-freedom structure using tuned viscous mass damper: the tuned viscous mass damper. Earthq. Eng. Struct. Dyn. 41(3), 453–474 (2012) 8. Hu, Y., Chen, M.Z.Q.: Performance evaluation for inerter-based dynamic vibration absorbers. Int. J. Mech. Sci. 99, 297–307 (2015) 9. Lazar, I.F., Neild, S.A., Wagg, D.J.: Using an inerter-based device for structural vibration suppression: using an inerter-based device for structural vibration suppression. Earthq. Eng. Struct. Dyn. 43(8), 1129–1147 (2014) 10. Garrido, H., Curadelli, O., Ambrosini, D.: Improvement of tuned mass damper by using rotational inertia through tuned viscous mass damper. Eng. Struct. 56, 2149–2153 (2013)

Chapter 12

Temperature Effects on the Modal Properties of a Suspension Bridge Etienne Cheynet, Jonas Snæbjörnsson, and Jasna Bogunovi´c Jakobsen

Abstract The paper studies temperature effects on the modal parameters of a suspension bridge across a Norwegian fjord. The approach used is a full-scale ambient vibration testing, where an automated Covariance-Driven Stochastic Subspace Identification (SSI-COV) method is used to identify the modal parameters. The bridge site, the bridge structure and the monitoring system are presented, followed by a summary of the data analysis procedure and the parameters used for the automated SSI-COV method applied. The operational modal analysis is based on 6 months of continuous acceleration records providing seasonal and diurnal variations of the natural frequencies of the bridge and the modal damping ratios. Temperature effects were observed with details that are scarcely available in the literature. In particular, the pronounced daily fluctuations of natural frequencies and seasonal effects are documented. Keywords Suspension bridge • Full-scale • Ambient vibrations • Modal parameters • Temperature effects

12.1 Introduction Ambient Vibration Testing (AVT) has become the “default procedure” for modal parameter identification of cable-supported bridges [3]. AVT is particularly attractive for studying environmental effects on the modal parameters, such as the evolution of the modal damping ratios of suspension bridge with the mean wind velocity [2, 4, 9, 13]. Other environmental effects, such as daily and seasonal fluctuations of temperature, are also known to influence the eigen-frequencies of concrete bridges [11, 14] and suspension bridges [6, 8]. During the last few years, several studies have focused on modelling the thermal behaviour of the entire bridge structure [5, 18, 22] and on measuring the static displacements induced by thermal loading [20, 21]. On the other hand, relatively few studies have been dedicated to the investigation of temperature effects on the modal properties of suspension bridges using AVT, based on large amount of ambient vibration data. The present paper illustrates the influence of daily temperature fluctuations on the eigen frequencies and modal damping ratios of the Lysefjord bridge, which is located at the inlet of a fjord in Norway. The automated Covariance-Driven Stochastic Subspace Identification (SSI-COV) algorithm developed by Magalhães et al. [11] has been used to identify the modal parameters of the bridge. This algorithm was applied by e.g. Magalhães and Cunha [10] on an arch bridge using more than 2 years of data and by Brownjohn et al. [3] on a long-span suspension bridge using only a couple of days of data. The present study therefore complements the study of Magalhães and Cunha [10] and Brownjohn et al. [3] by utilising 6 months of continuous vibration measurements conducted on a long-span suspension bridge. The present paper is organised as follows: first, the bridge site and instrumentation are presented, followed by a short summary of the parameters used with the automated SSI-COV algorithm. Then the influence of the daily temperature fluctuations on the bridge eigen-frequencies is demonstrated and discussed. Finally, the evolution of the modal damping ratios with temperature variations are investigated.

E. Cheynet () • J.B. Jakobsen Department of Mechanical and Structural Engineering and Materials Science, University of Stavanger, N-4036 Stavanger, Norway e-mail: [email protected] J. Snæbjörnsson Department of Mechanical and Structural Engineering and Materials Science, University of Stavanger, N-4036 Stavanger, Norway School of Science and Engineering, Reykjavik University, Menntavegur 1, 101 Reykjavík, Iceland © The Society for Experimental Mechanics, Inc. 2017 J. Caicedo, S. Pakzad (eds.), Dynamics of Civil Structures, Volume 2, Conference Proceedings of the Society for Experimental Mechanics Series, DOI 10.1007/978-3-319-54777-0_12

87

88

E. Cheynet et al.

12.2 Instrumentation and Methods 12.2.1 The Bridge Site The Lysefjord Bridge, positioned at the narrow inlet of a fjord in the South-West part of the Norwegian coast, is used as a case study. Its main span is 446 m, with the central part 55 m above the sea level. The bridge is entrenched between two steep hills with slopes ranging from 30 to 45° and a maximum altitude of 350 m to the North and 600 m to the South. The bridge is exposed to winds that may descent from mountains nearby or follow the fjord over a longer path. The West side of the bridge is exposed to a more open and levelled area. Since 2013, the bridge has been instrumented with seven sonic anemometers installed on the West side of the bridge deck and four pairs of accelerometers located inside the deck (Fig. 12.1). GPS timing is used to synchronize the windand acceleration data, acquired locally with separate data logging units. Although the highest sampling frequency of the accelerometers and anemometers is 200 Hz and 32 Hz respectively, the data are acquired with a sampling frequency of 50 Hz and decimated to 20 Hz to facilitate data handling. Finally, a 3G router enables wireless data access and transfer via a mobile network. A more detailed description of the bridge instrumentation is given in e.g. [4]. Temperature measurements are provided by a Vaisala weather transmitter WXT520 located on hanger 10, denoted H-10, whereas the accelerometers are located near hangers 9, 18, 24 and 30, where hanger 18 corresponds to the mid-span position.

12.2.2 Automated SSI-COV Procedure The automated SSI-COV method developed by Magalhães et al. [11] is applied on 6 months of acceleration records obtained from July 2015 to December 2015. For the sake of brevity, the automated SSI-COV method used will not be described explicitly in this paper but the details can be found in e.g. [10, 11]. The parameters used to calibrate this automated SSICOV method are summarized in Table 12.1. The minimal and maximal order of the system for the calculation of the stabilization diagram are denoted Nmin and Nmax respectively, whereas max is the maximal time lag used to compute the cross-covariance matrix. The three accuracy thresholds for the identified eigen-frequencies, modal damping ratios and modal assurance criterion [1] are denoted fn ,  and MAC respectively. Finally, the threshold accuracy for the cluster analysis is d . The mode shapes and eigen-frequencies of the Lysefjord Bridge were successfully identified by the automated SSI-COV method used in the present study [4]. The modal parameters are hereby denoted using the code XYZ, where X D fH; V; Tg represents the lateral (H), vertical (V) and torsional (T) bridge motion. Y D fS; Ag is the symmetric (S) or asymmetric (A) mode shape, and Z the mode number. For example HS1 refers to the first symmetric horizontal mode shape, and TA2 refers to

Accelerometer Measuring GNSS Reference GNSS

x y

N

E

W

S

z

Sonic anemometer 127 m

72 m

North tower

24 m 24 m

H-16 H-08 H-09 H-10

48 m

H-30

H-20

South tower

H-24

H-18

Fig. 12.1 Wind and structural health monitoring system installed on the Lysefjord Bridge Table 12.1 Parameters used in the SSI-COV method applied on Lysefjord bridge acceleration data max (s) 15

Nmin 3

Nmax 30

fn 5e3

 3e2

MAC 5e3

d 2e2

12 Temperature Effects on the Modal Properties of a Suspension Bridge Table 12.2 Eigen-frequencies estimated from the measurement data using the SSI-COV algorithm in comparison to the values based on the SBM method and the Abaqus model

89

Modes HS1 HA1 HS2 HA2 HS3 HA3 VA1 VS1 VS2 VA2 VS3 VA3 TS1 TA1

SSI-COV Hz 0.136 0.444 0.577 0.626 0.742 1.011 0.223 0.294 0.408 0.587 0.853 1.163 1.237 2.184

SBM Hz 0.130 0.442 0.556 0.597 0.830 1.000 0.205 0.319 0.439 0.585 0.864 1.194 1.215 2.186

% 4.41 0.45 3.51 4.61 11.90 1.03 8.10 8.35 7.63 0.39 1.31 2.72 1.78 0.09

Abaqus Hz % 0.128 6.19 0.431 2.90 0.533 7.56 0.583 6.81 0.833 12.31 0.974 3.69 0.214 3.91 0.302 2.72 0.407 0.25 0.583 0.68 0.856 0.34 1.191 2.36 1.238 0.026 2.122 2.85

the second asymmetric torsional mode shape. To increase the identification speed of the lower modes, the sampling frequency of the lateral and vertical acceleration records were reduced to 2 Hz. The sampling frequency of the torsional acceleration response remained at 20 Hz. This allowed the SSI-COV algorithm to be applied to more than 50;000 acceleration records of 10 min duration in less than half a day. A Finite element (FE) model created by Steigen [15], using the Abaqus software, and improved by Tveiten [17] was used to evaluate a numerical prediction of the mode shapes and eigen-frequencies of the Lysefjord Bridge. The eigenfrequencies and the mode shapes of the Lysefjord Bridge were also approximated by using harmonic series expansions following Sigbjörnsson and Hjorth-Hansen [12] for the lateral motion and Strømmen [16] for the vertical and torsional ones. In the following, the latter method is referred to as the “Simplified Bridge Model” denoted SBM. The eigen-frequencies evaluated using the two different models and the SSI-COV algorithm are listed in Table 12.2.

12.3 Results 12.3.1 Influence of Temperature Variations on the Eigen-Frequencies As stated by Xia et al. [19], a higher temperature leads in general to decreased values of vibration frequencies, mainly due to the temperature dependency of the materials Young’s modulus. Such variations of the eigen-frequencies are visible in Fig. 12.2, except for temperatures over 20 ıC where the number of samples was probably too low to provide reliable results. The influence of temperature variations on the first lateral mode HS1 is rather small. The frequency drops from 0.139 to 0.135 Hz when the temperature increases from 0 to 20 ıC. For a similar temperature change, the frequency associated with VA1 decreases only from 0.227 to 0.220 Hz. The most dramatic frequency change occurs for the mode TS1 where the frequency decreases from 1.25 to 1.23 Hz. The scatter of the eigen-frequencies observed on Fig. 12.2 is due to the influence of other parameters such as traffic and wind excitation. The daily fluctuations of the eigen-frequencies can be visualized by studying few days of data. This is done in Fig. 12.3, where 10 days of data recorded in October 2015 are displayed. The first lateral eigen-frequency HS1 fluctuates between 0.132 Hz for diurnal data and 0.145 Hz for nocturnal data. These fluctuations are relatively small compared to those from VA1 which ranges from 0.217 Hz during day time to almost 0.230 Hz during the night. For the torsional motion, TS1 fluctuates between 1.25 Hz down to 1.23 Hz. As suggested by Kim et al. [7], heavy traffic is likely to be responsible for a decrease in the estimated eigen-frequencies of the bridge deck. The effects of the temperature and traffic on the bridge eigen-frequencies are therefore expected to superimpose and be responsible for larger frequency variations. At night time, the lower temperature and the reduced traffic leads to higher eigen-frequencies whereas at day time, the increase of the temperature and traffic leads to lower eigenfrequencies. This appears clearly on Fig. 12.3, where a pseudo-period of 24 h is visible. The periodical pattern is clearly visible for the vertical bridge motion, but it is more noisy for the lateral and torsional motions. This can be partly explained by the higher signal to noise ratio measured for the vertical motion.

90

E. Cheynet et al.

f (Hz)

0.14

0.46 0.45

0.135 0.44

f (Hz)

(HS1) 0.13

0.43

0.59

0.64

0.58

0.63

f (Hz)

(HS2)

(HA2)

0.57

0.62

0.3

0.23

0.295

0.223

f (Hz)

(VS1)

(VA1)

0.29

0.215

0.42

0.6 0.59

0.41 (VS2)

f (Hz)

(HA1)

(VA2)

0.4

0.58

1.26

2.21

1.24

2.18 (TA1)

(TS1) 2.16

1.22 5

10

15

20

5

10

T(°C)

15

20

T(°C)

Fig. 12.2 Evolution of the first four lateral and vertical and the first two torsional eigen-frequencies with the temperature. The data set comprises 6 months of acceleration and temperature records (July–December 2015)

Temperature fluctuations seem to have a larger influence on the variation of the eigen-frequencies than the traffic loading. The attenuation of the daily periodicity of the eigen-frequency in November and December (Fig. 12.4) cannot simply be explained by a reduction of heavy traffic for example. The periodicity pattern appears to be almost entirely modulated by temperature changes. For example, we observed that the sinusoidal pattern was elongated at the bottom part in July (longer day) but elongated at the top part in October (shorter day), without strong variations in the amplitude of the fluctuations.

12.3.2 Influence of Temperature Variations on the Modal Damping Ratios The estimation of the modal damping ratios is one of the most crucial step in studying accurately the buffeting response of a suspension bridge. Unfortunately, such studies are a rarity in full scale. In general, the aerodynamic damping ratios are obtained with a large dispersion in full-scale [2, 13]. This requires a statistically significant amount of data, which is rarely presented in the literature. In this subsection, the total damping is considered for various wind conditions, using a considerable amount of data. The evolution of the modal damping ratios with the mean wind velocity has been described in e.g. Cheynet et al. [4] and is therefore not recalled here. Temperature effects on the modal damping ratios remain mostly unexplored and are therefore briefly investigated in the following. The variation of the modal damping ratios with temperature is displayed in Fig. 12.5, for the first four lateral and vertical modes as well as the first two torsional modes.

12 Temperature Effects on the Modal Properties of a Suspension Bridge

f (Hz)

0.15 0.145 0.14 0.135 0.13

Mean + RMS

Mean

91

aw data

f (Hz)

0.23 0.225 0.22

f (Hz)

0.215 1.26 1.25 1.24 1.23 1.22 10/10

12/10

14/10 16/10 Time (DD/MM)

18/10

20/10

Fig. 12.3 Evolution of HS1 (top), VA1 (middle) and TS1 (bottom) between the 20/09/2015 and 30/09/2015. Data binning has been applied to better estimate the fluctuating mean value and RMS of the eigen-frequencies

f (Hz)

(VA1)

20

10

T (◦C)

0.23

0.22 0 01/07

01/08

01/09

01/10

01/11

01/12

Time (DD/MM) Fig. 12.4 Evolution of the temperature and the frequency of the mode VA1 from July 2015 to December 2015

As expected, the damping ratios for HS1, HS2, HA2, VS1, VA1 and TA1 fluctuate little with the temperature. For a given temperature, multiple wind velocities are recorded. Because the total modal damping ratios increase with the wind velocity, due to added aerodynamic damping, a large variety of total modal damping ratios are recorded for a given temperature. Consequently, a large dispersion is expected when the modal damping ratios are expressed as a function of the temperature alone. On Fig. 12.5, VS1 and VA1 have an averaged modal damping ratios of about 0.75% and 1% respectively and are characterized by a considerable spreading. This is not surprising given that the modal damping ratios associated with VS1 and VA1 are highly affected by the wind induced aerodynamic damping effect [4]. A reduction in modal damping with increasing temperatures and the underlying reduction in wind speed is generally expected. This is in particular the case for the vertical modes for which the aerodynamic damping can be relatively high. The highest damping values are still observed for temperatures between 5 ıC and 10 ıC at which the strongest winds occur. The decrease in the damping ratios of TS1 for increasing temperatures is somewhat more distinct. As already pointed out, lower temperatures are typically associated with higher wind speeds, for which a weakly increasing TS1 damping ratio has been observed [4]. It is however a bit surprising that the corresponding contribution of the aerodynamic damping, which is much stronger for the vertical modes, is not so obvious in the average damping estimates presented. To improve the reliability of these results, an analysis using at least 1 year of temperature, velocity and acceleration data is required.

92

E. Cheynet et al.

4 3 2 1 0

4 3 2 1 (HS2) 0

4 3 2 1 0

2 1.5 1 0.5 (VS1) 0

2 1.5 1 0.5 0

2 1.5 1 0.5 (VS2) 0

2 1.5 1 0.5 0

2 1.5 1 0.5 (TS1) 0

2 1.5 1 0.5 0

ζ (%)

ζ (%)

ζ (%)

ζ (%)

ζ (%)

4 3 2 1 (HS1) 0

5

10

15 T(°C)

20

(HA1)

(HA2)

(VA1)

(VA2)

(TA1) 5

10

15

20

T(°C)

Fig. 12.5 Modal damping ratios expressed as a function of the temperature for the first four lateral and vertical modes and the first two torsional modes, based on acceleration data recorded from July 2015 to December 2015

12.4 Conclusions The modal parameters of the Lysefjord bridge have been identified using AVT based on an automated SSI-COV procedure relying on several months of continuous acceleration records. Environmental effects were observed with a level of details that is scarcely available in the literature. In particular, the daily fluctuations of the eigen-frequencies were remarkably well captured as was the possible temperature-dependency of the modal damping ratios. A relatively large amount of acceleration data was accumulated, so that a statistical description of the influence of the mean wind velocity on the modal damping ratios could be achieved. The application of OMA using the automated SSI-COV method on other suspension bridges may provide a better general understanding of environmental influence on the modal parameters of such structures. The good agreement between the computed natural frequencies and the measured ones is encouraging for further investigation of full-scale measurement data using the automated SSI-COV method. The modal analysis also documented the non-linearity of the modal damping ratios and a possible non-negligible role of temperature effects on the modal parameters of the bridge. Further analysis will consider at least 1 year of full-scale measurements, and a more severe segregation of wind samples characterized by unusually high turbulence intensity indicating non-stationary fluctuations in excitation and response. Acknowledgment The authors are grateful to the Norwegian Public Road Administration for the support of and the assistance during the measurement campaign at the Lysefjord Bridge.

12 Temperature Effects on the Modal Properties of a Suspension Bridge

93

References 1. Allemang, R.J., Brown, D.L.: A correlation coefficient for modal vector analysis. In: Proceedings of the 1st International Modal Analysis Conference, SEM, Orlando, vol. 1, pp. 110–116 (1982) 2. Brownjohn, J.M.W., Dumanoglu, A.A., Severn, R.T., Taylor, C.A.: Ambient vibration measurements of the Humber Suspension Bridge and comparison with calculated characteristics. Proc. Inst. Civil Eng. 83(3), 561–600 (1987) 3. Brownjohn, J.M.W., Magalhaes, F., Caetano, E., Cunha, A.: Ambient vibration re-testing and operational modal analysis of the Humber Bridge. Eng. Struct. 32(8), 2003–2018 (2010). doi:10.1016/ j.engstruct. 2010.02.034 4. Cheynet, E., Bogunovic’ Jakobsen, J., Snæbjörnsson, J.: Buffeting response of a suspension bridge in complex terrain. Eng. Struct. 128, 474–487 (2016). doi:10.1016/ j.engstruct.2016.09.060 5. de Battista, N., Brownjohn, J.M.W., Pink Tan, H., Koo, K.-Y.: Measuring and modelling the thermal performance of the Tamar Suspension Bridge using a wireless sensor network. Struct. Infrastruct. Eng. 11(2), 176–193 (2015). doi:10.1080/ 15732479.2013.862727 6. Ding, Y., Li, A.: Temperature-induced variations of measured modal frequencies of steel box girder for a long-span suspension bridge. Int. J. Steel Struct. 11(2), 145–155 (2011). doi:10.1007/ s13296- 01120044 7. Kim, C.-Y., Jung, D.-S., Kim, N.-S., Kwon, S.-D., Feng, M.Q.: Effect of vehicle weight on natural frequencies of bridges measured from traffic-induced vibration. Earthquake Eng. Eng. Vib. 2(1), 109–115 (2003). doi:10.1007/ BF02857543 8. Koo, K.Y., Brownjohn, J.M.W., List, D.I., Cole, R.: Structural health monitoring of the Tamar suspension bridge. Struct. Control Health Monit. 20(4), 609–625 (2013) 9. Macdonald, J.H.G.: Evaluation of buffeting predictions of a cable-stayed bridge from full-scale measurements. J. Wind Eng. Ind. Aerodyn. 91(12–15), 1465–1483 (2003). doi:10.1016/ j.jweia.2003.09.009. ISSN:0167-6105 10. Magalhães, F., Cunha, À.: Explaining operational modal analysis with data from an arch bridge. Mech. Syst. Signal Process. 25(5), 1431–1450 (2011). doi:10.1016/ j.ymssp.2010.08.001. ISSN: 0888-3270 11. Magalhães, F., Cunha, À., Caetano, E.: Online automatic identification of the modal parameters of a long span arch bridge. Mech. Syst. Signal Process. 23(2), 316–329 (2009). doi:10.1016/ j.ymssp.2008.05.003. ISSN: 0888-3270 12. Sigbjörnsson, R., Hjorth-Hansen, E.: Along-wind response of suspension bridges with special reference to stiffening by horizontal cables. Eng. Struct. 3(1), 27–37 (1981) 13. Siringoringo, D.M., Fujino, Y.: System identification of suspension bridge from ambient vibration response. Eng. Struct. 30(2), 462–477 (2008). doi:10.1016/ j.ymssp.2008.05.003 14. Sohn, H., Dzwonczyk, M., Straser, E.G., Kiremidjian, A.S., Law, K.H., Meng, T.: An experimental study of temperature effect on modal parameters of the Alamosa Canyon Bridge. Earthquake Eng. Struct. Dyn. 28(8), 879–897 (1999). ISSN: 1096-9845 15. Steigen, R.O.: Modeling and analyzing a suspension bridge in light of deterioration of the main cable wires. MA thesis. University of Stavanger (2011) 16. Strømmen, E.N.: Eigenvalue calculations of continuous systems. In: Structural Dynamics, pp. 89–159. Springer International Publishing, Cham (2014). doi: 10.1007/ 9783319018027_3. ISBN: 978-3-319-01802-7 17. Tveiten, J.: Dynamic analysis of a suspension bridge. MA thesis. University of Stavanger (2012) 18. Westgate, R., Koo, K.-Y., Brownjohn, J.: Effect of solar radiation on suspension bridge performance. J. Bridge Eng. 20(5), 04014077 (2014). doi:10.1061/ (ASCE)BE.19435592.0000668 19. Xia, Y., Chen, B., Weng, S., Ni, Y.Q., Xu, Y.L.: Temperature effect on vibration properties of civil structures: a literature review and case studies. J. Civ. Struct. Health Monit. 2(1), 29–46 (2012). doi:10.1007/ s13349- 01100157 20. Xia, Y., Chen, B., Zhou, X.Q., Xu, Y.L.: Field monitoring and numerical analysis of Tsing Ma Suspension Bridge temperature behavior. Struct. Control Health Monit. 20(4), 560–575 (2013). doi:10.1002/ stc.515 21. Xu, Y.L., Chen, B., Ng, C.L., Wong, K.Y., Chan, W.Y.: Monitoring temperature effect on a long suspension bridge. Struct. Control Health Monit. 17(6), 632–653 (2010). doi:10.1002/ stc.340 22. Zhou, L., Xia, Y., Brownjohn, J.M.W., Young Koo, K.: Temperature analysis of a long-span suspension bridge based on field monitoring and numerical simulation. J. Bridge Eng. 21(1), 04015027 (2016). doi:10.1061/ (ASCE)BE.19435592.0000786

Chapter 13

Mass Scaling of Mode Shapes Based on the Effect of Traffic on Bridges: A Numerical Study M. Sheibani, A.H. Hadjian-Shahri, and A.K. Ghorbani-Tanha

Abstract In order to derive mass normalized mode shapes from Operational Modal Analysis (OMA) techniques, additional experiments have to be conducted in the interest of scaling the determined mode shapes. Various investigations have been carried out based on the deterministic perturbed mass matrix, also known as the mass change method. However, the conventional form of this method requires a number of rather costly and sometimes impractical experiments on the structure. In this article, it is intended to use traffic as a stochastic source of mass change to calculate mass scaling factors. Adequate traffic stream on bridges can affect the Eigen properties of the structure efficiently. The vehicles on the bridge are only considered to affect the mass properties, and vehicle-bridge interaction is neglected. A simplified structural model of a bridge is considered and the traffic is modeled as lognormal distribution. Nodal masses induced by the vehicles are converted to time histories to avoid difficulties based on the moving mass problem. Consequently, a method is proposed to produce the modified mass matrix which can be used in the scaling factor equations. Finally, scaling factors of mode shapes are proposed by comparing the unperturbed structure and the perturbed mass matrix structure. Keywords Mass change strategy • Mode shape scaling • Traffic induced • Lognormal distribution • Modified mass matrix

13.1 Introduction Operational Modal Analysis (OMA) techniques have been widely used during the recent decades and have nearly made Experimental Modal Analysis (EMA) obsolete. The advantages of OMA, such as elimination of the need for heavy shakers and the ability to perform tests without interruption of normal structure service, have made this method a reliable substitution for the EMA technique. However, certain drawbacks exist in every approach of OMA. As these methods consider the input force to be the unknown ambient noise, there is not a direct method available to normalize the extracted mode shapes and additional steps need to be taken [1]. Mass scaled modes are essential in numerous applications of modal analysis such as structural response simulation, damage detection, health monitoring applications, model updating, etc. [2, 3]. Several methods have been proposed to overcome this issue and they all share the notion that a controlled perturbation to the dynamics of the structure is the key element in obtaining scale factors of the mode shapes. The most promising methods have been found to be mass change techniques, in which known masses are added to the structure and the tests are performed before and after the addition of masses. The scale factors, consequently, can then be calculated by comparing the results. Parloo et al. proposed a sensitivity based method for obtaining scale factors of each mode shape. Mass change method was first introduced and validated by comparing the scale factors derived from forced vibration test, with those from repeated in-operation test considering different locations of the added mass [4]. Parloo et al. further evaluated the normalization method by employing it in a full-scale bridge test. Heavy concrete blocks were added to specific locations of the bridge in order to provide sufficient frequency shifts that are desired for the mass change method [5]. Brinker and Andersen further studied the method and derived a formula based on the equation of motion to estimate the scale factors [6]. Other studies investigated the best mass change setups and sources of error [2, 7]. Aenlle et al. offered optimized strategies regarding the method and studied the best mass ratios, optimized locations and uncertainties relating the method. It was shown that the best mass change scenarios were those which induced uniform mass changes to the entire structure and consequently prevented alteration of the mode shapes. Furthermore, the ratios of added masses to the entire structure were recommended to be high

M. Sheibani () • A.H. Hadjian-Shahri • A.K. Ghorbani-Tanha School of Civil Engineering, College of Engineering, University of Tehran, P.O. Box: 11155-4563, Tehran, Iran e-mail: [email protected] © The Society for Experimental Mechanics, Inc. 2017 J. Caicedo, S. Pakzad (eds.), Dynamics of Civil Structures, Volume 2, Conference Proceedings of the Society for Experimental Mechanics Series, DOI 10.1007/978-3-319-54777-0_13

95

96

M. Sheibani et al.

enough to produce frequency shifts desired for the method, and low enough to avoid significant changes of mode shapes [8]. Mass-stiffness change method was also proposed to change the first modes of vibration more significantly [9]. The growing accuracy of modal analysis testing equipment and precise methods which have made obtaining the modal parameters possible, as well as the difficulties associated with placing heavy masses on the structure when using the conventional mass change methods, has revealed the need for more convenient and efficient methods. In this article a new method is proposed to exploit the traffic load as the required change in the mass of the structures. The heavy traffic jams or congested traffic situations observed in big cities has appealing characteristics which makes it useful for utilization in the mass change method. Vehicle traffic induces a sufficiently uniform load on the bridges in the case of congested traffic and this condition can be assumed almost time independent for the required time period concerning OMA. Equally distributed mass change strategy is suggested to be the optimum approach in the literature and is provided by the traffic load. On the other hand, there are various urban bridges which are constructed using light-weight materials that promise sufficient frequency shifts desired for the mass change method provided by the traffic stream. Near-congestion traffic has been shown to be best represented by shifted lognormal distribution [10, 11]. To demonstrate, a bridge-like structure is simulated by a finite-element model of a simply-supported Euler-Bernoulli beam. The beam model has been subjected to artificial traffic generated from lognormal distribution and the proposed method has been evaluated. Mass change is applied to the mass matrix of the beam in each time-step to represent the actual case of traffic, and responses of the structure has been determined by the Newmark method. The car-bridge interaction is ignored since vehicles in congested traffic conditions have negligible dynamic interactions with bridges, and therefore a decoupled bridge-vehicle system is considered [12]. The response-only analysis method that is used in this study is the Natural Excitation Technique-Eigensystem Realization Algorithm (NExT-ERA) which has shown promising results [13] and can be exploited for repetitive tests needed in this study. Unscaled mode shapes are derived from response-only modal analysis and the corresponding scale factors are calculated based on the traffic induced mass modification. These scale factors are then compared with the scale factors obtained from finite-element model.

13.2 Theory 13.2.1 Output-Only Modal Identification The method which is used for modal analysis purposes is NExT-ERA. The prevailing assumptions made in this method are that the structure behaves within a linear range, the structure is time-invariant, and the forces applied to the system are uncorrelated to the response of the structure [15]. In this method, response of the structure based on ambient excitation is used for estimating the cross-correlation functions. These functions can represent the impulse response functions of the structure which are used in classical modal analysis. The direct procedure for obtaining the cross-correlation functions between two channels of acceleration i and j is used to this end. In discrete time approach, the correlation function (CF) matrix can be estimated using the formula [14] RyRi ;Ryj .kt/ D

Sk 1 X yR i .s/Ryj .s C k/ N  k sD1

(13.1)

In which S is the total number of data points of the acceleration record and t is the time step. The ERA method is based on the state-space representation of a discrete system. The procedure to obtaining modal parameters from ERA method starts with formation of the Hankel matrix using correlation functions which are calculated by Eq. (13.1). Singular value decomposition (SVD) is then performed and modal parameters are derived with the help of specific equations and guidelines which are described in [15, 16].

13.2.2 Mass Scaling In order to obtain mass normalized mode shapes from OMA techniques, additional steps are required. The unscaled mode shape vector § i have to be scaled in the interest of modal analysis applications mentioned in the introduction. The scaled mode shapes can be obtained from unscaled modes using the following equation [8]

13 Mass Scaling of Mode Shapes Based on the Effect of Traffic on Bridges: A Numerical Study

®i D ˛i :§ i

97

(13.2)

where ˛ i is the scaling factor of mode shape i. The proposed methods existing in the literature for determining the scaling factors, are based on the eigenvalue changes of the structure before and after the addition of the masses. The eigenvalue problem of equation of motion in the absence of damping is m:®0 :!02 D k:®0

(13.3)

where m and k are mass and stiffness matrices of the structure, respectively, ®0 is the mass normalized mode shape and ! 0 is the natural frequency. Added masses are often considered lumped masses for simplification purposes and thus the mass change matrix m becomes in general diagonal. The eigenvalue problem of equation of motion in the presence of additional masses is in the following form .m C m/ :®1 :!12 D k:®1

(13.4)

in which ®1 and ! 1 are modal parameters of the modified structure. It has been shown that the scaling factors of unscaled mode shapes can be derived from the following equation [8] s  ˛D

!02  !12



!12 :§ T :m:§

(13.5)

However, according to [3], if a full set of modes is used, the exact scale factors can be derived from the following equation and this equation is fulfilled for each value of i v u u u ˛j D t

  !0j2  !1i2 :Bji   § 0j T : !1i2 :m :§ 1i

(13.6)

in which matrix B can be produced by the equation b b 1 BD§ 0 :§ 1

(13.7)

1

b 0 is the pseudo-inverse of § 0 . where §

13.2.3 Modeling the Traffic Excitation Traffic load is shared between nodes of the system using a physics-based traffic excitation model. In this method, every vehicle induces a moving load when it is traversing the bridge and this load can be assumed to affect each node of the structure with a time history load [17]. Concentrated and constant vehicle forces are represented by P during vehicle movement over the bridge. The governing equation of motion for vertical deflection of the bridge is m

@2 @ @4 .x; .x; y t/ C t/ C y c EI yb .x; t/ D ı .x  vt/ P b b @t2 @t @x4

(13.8)

in which, m, c, E and I are the mass per unit length, the damping coefficient, the Young’s modulus and the cross-sectional moment of inertia of the beam, respectively, and v is the velocity of the vehicle while yb is the vertical deflection of the bridge from the equilibrium position. Horizontal position of the forces is shown by x as illustrated in Fig. 13.1. Considering the forces induced by multiple vehicles on the bridge as a summation and assuming a lognormal distribution for the headways of the successive arrivals of the vehicles, one can replace the right hand side of Eq. (13.8) by

98

M. Sheibani et al.

Fig. 13.1 Model of moving load on the bridge [18]

Fx .t/ D 

N.t/ X

Pi ı Œx  v .t  i /

(13.9)

iD1

where Fx (t) is the time history of traffic excitation at location x on the bridge, N(t) is the number of vehicles on the bridge at the time t, and 1 , 2 , . . . i , . . . N are the sequence of headway times. The shifted lognormal distribution is used to model headway times in various studies and in this study is used to generate headway times of the artificial traffic [10]. The probability distribution equation is .ln .t  ˇ/  /2 f .tjˇ; ; / D  exp  p 2 2

.t  ˇ/ 2 1

! It >

(13.10)

in which t is the time headway, ˇ is the shift value in seconds, and  and are parameters of lognormal distribution; location and scale parameter, respectively. Dynamic nodal loading (DNL) is used to convert time-varying moving forces into load histories at each node of the Finite Element Method (FEM) model based on the equivalent nodal forces (ENFs) method [18]. The bridge is modeled by beam elements and each node of the beam has two degrees of freedom; yi , the transvers displacement and  i , the in-plane rotation. The vertical force P is then applied to nodes of the structure as nodal shear WiQ and nodal moment WiM using the following equations 8 ˆ ˆ h0 i ˆ ˆ .li1 xi Cvt/2 2.xi vt/ < 1 C 2 li1 li1 h i WiQ .t/ D .li Cxi vt/2 2.vtxi / ˆ 1 C ˆ 2 ˆ l i li ˆ : 0 8 0 ˆ ˆ ˆ 2 ˆ i Cvt/ <  .li1 x .xi  vt/ 2 li1 M Wi .t/ D 2 .li Cxi vt/ ˆ .vt  xi / ˆ ˆ li2 ˆ : 0

  xi li1 t  v   xi li1 xi v  xi v < t x i < t  i Cl v v xi Cli $1,500,000

232

S. Vemuganti et al.

For preliminary layout of the quantification of impacts and consequences using objective data, the following general scenario is proposed: With truck speeds of 20 km/h, the permanent deformation in the model is observed to be 2.5 mm. For such small deformations, the bridge will still remain in operational condition with no negligible damage. Therefore, this impact is classified as category 1. Due to the increased momentum of vehicle in 40 km/h skewed and non-skewed impact cases, higher deformations are observed in the model. The elevated deformation increases the severity of the collision to category 2. Thus, this bridge would experience some temporary service interruptions such as train speed reduction. A significant permanent deformation of 75 mm is experienced by the bridge in the impact case with truck speed of 60 km/h. With such nonlinearities, the bridge department would have to avoid potential service interruptions, causing permanent train speed reductions until the repair. The severity of this case has therefore been classified as category 4 in the rating system. The trends observed from the results of the numerical simulations indicate that higher velocity impacts have the potential to create impact severity of category 5. These proposed thresholds are the first preliminary effort to “rate” measured consequences with current railroad bridge management decisions. For the application of this theoretical effort to railroad environments and decisions, future research will be conducted in coordination with the railroad departments input, including workshops and small scale tests.

28.5 Future Direction: Spatial Identification of Impact Damage In order to evaluate the severity of a collision after the impact event, the energy inserted into the system must be identified. The level of energy depends on variables such as the speed of the truck and the location of impact. Based on previous studies, which demonstrate the potential of neural networks for detection and characterization of nonlinearities in dynamic systems [21, 22], an effective and simple approach for identification of damage location of bridge-vehicle collisions is proposed here. To demonstrate the identification procedure and represent the dynamic response of a TPG, a simplified FE model of a fixed-pinned beam is considered. Since this response is governed by the six modes, the beam model consists of six elements, where each node has a translational and rotational degree of freedom. The beam has properties representative of the three-dimensional bridge considered above. This has been achieved by adjusting the mass, stiffness and damping matrix. The damping matrix has been computed as Rayleigh damping proportional to the mass and stiffness matrix, respectively. The modal damping ratio of the first mode is tuned to 0.02. The aim is to detect the spatial location of damage. Therefore, the model has been split into six sections as illustrated in Fig. 28.7. Each of the six elements is given stiffness properties denoted by k, with the subscripts 1–6. Damage caused by an impact event is assumed to cause local stiffness reduction. This has been simulated numerically by weakening the stiffness of

Fig. 28.7 Simplified beam model of the railway bridge used for the neural networks approach

28 Sensing and Rating of Vehicle–Railroad Bridge Collision

233

Fig. 28.8 (a) Beam model of the bridge span, where the red element indicates stiffness reduction. (b) The probability of the damage location found by the proposed procedure Table 28.3 Selection of damage sensitive features of the beam model

Feature 1 2 3 4 5 6 7 8

Damage sensitive features Difference between displacement time history at reference location Root-mean-square of reference response Correlation coefficient of reference response Mean of reference response Standard deviation of reference response Natural frequency, f1 and f2 Modal damping ratio,  1 and  2 Euclidian norm of normalized mode shape, jj 1 jj2 and jj 2 jj2

subsequent elements. The initial step of the identification is to select the damage sensitive features. These are listed in Table 28.3 and consist of statistical quantities achieved from the dynamic response of the numerical model. The dynamic response is obtained from the two reference signals from the locations indicated in Fig. 28.7. Furthermore, the modal parameters for the first and second mode are selected as features. The patterns of the features selected in response to stiffness reduction have been classified to train and validate the neural network. For training, 230 data sets are simulated by individually reducing the stiffness of k1 –k6 by 10, 20, 30, 40, 50, 60, 70, 80 and 90%. The data set contains the features obtained from the impact at the middle of the beam. Forty-nine samples each are employed for validation and testing. To improve training of the neural network, a feature space reduction step is performed. The space reduction is achieved by applying principal component analysis as in [22]. It is desirable to have the largest possible data set for training of neural networks where the size of the data set grows with an increasing number of features. The training is performed using MATLAB Neural Network Toolbox with nine neurons. A single neuron act as a correspondence to the input-output mapping by regression. The extension to multiple neurons is achieved by establishing a neural network, such that the output of a neuron can be the input of another [22]. The training of the network has been performed to obtain weighting functions that correlate to the features with the damage location. After completion of training, an additional data set has been generated. This data set has been used to test the robustness of the identification procedure. The stiffness in the middle region of the beam has been decreased by 19.96% instead of 20% as in the training data set, shown in Fig. 28.8a. The outcome probability for the damage to be in the middle of the beam model is 100%, as given in Fig. 28.8b. Other combinations of stiffness reduction of the beam elements have also been tested. The true location has been consistently identified with a probability ranging from 80 to 100%. This indicates that the proposed procedure for the spatial identification of damage is reasonable for the developed numerical model. Similar procedure can be adopted for a detailed numerical model, such as the above ANSYS model. This would in particular require an extensive dataset, containing features as proposed in Table 28.3. The extent of this dataset depends on which accuracy of stiffness reduction due to vehicle impact is sought.

28.6 Conclusions and Future Work This paper emphasizes the type of accidents occurring due to impact between highway trucks and railway bridges. In order to study the effects of these impacts and rate them, we developed: (1) a preliminary FE model to understand the deformation processes associated with an impact event, and (2) an idealized neural network model to demonstrate its applicability in

234

S. Vemuganti et al.

detecting damage. The results presented herein represent an initial attempt to quantify truck-bridge impact phenomena and develop techniques to detect and mitigate future impact events. In our future work, the FE model will be improved by developing a higher fidelity mesh to more accurately capture the internal mechanical behavior of the bridge. In addition, we will develop a more accurate representation of a typical semi-truck. A mesh convergence study will be conducted in order to obtain an accurate and computationally efficient model. Furthermore, a more complex neural network system will be developed to rate the impacts when railway bridges are hit by highway traffic. A hybrid approach will be considered where both the location and the magnitude of the impact detected by neural networks are utilized as inputs to the simulation. Accordingly, structural integrity will be assessed using hybrid simulation results. Additionally, permanent displacements at the supports due to an impact will be evaluated. Finally, the effectiveness of crash beams in protecting the bridge against impacts will be investigated. Acknowledgment The authors of this paper thank the Canadian National Railway and the Canadian Pacific for their help in the development of this research methodology. The authors also thank Duane Otter from the Transportation Technology Center, Inc. TTCI), a wholly owned subsidiary of the Association of American Railroads (AAR) for his constructive feedback and recommendations.

References 1. Otter, D., Joy, R., Jones, M., Maal, L.: Need for bridge monitoring systems to counter railroad bridge service interruptions. Transp. Res. Rec. J. Transp. Res. Board. 2313(1), 134–143 (2012) 2. Joy, R., Jones, M.C., Otter, D., Maal, L.: Characterization of Railroad Bridge Service Interruptions. Railroad Bridges (No. DOT/FRA/ORD13/05), (2013) 3. Bischoff, R., Meyer, J., Enochsson, O., Feltrin, G., Elfgren, L.: Event-based strain monitoring on a railway bridge with a wireless sensor network. In: Proceedings of the 4th International Conference on Structural Health Monitoring of Intelligent Infrastructure, pp. 74–82. Zurich (2009) 4. Staszewski, W.J., Mahzan, S., Traynor, R.: Health monitoring of aerospace composite structures–Active and passive approach. Compos. Sci. Technol. 69(11), 1678–1685 (2009) 5. Farrar, C.R., Worden, K.: An introduction to structural health monitoring. Philos. Trans. R. Soc. Lond. A Math. Phys. Eng. Sci. 365(1851), 303–315 (2007) 6. Sohn, H.: Effects of environmental and operational variability on structural health monitoring. Philos. Trans. R. Soc. Lon. A Math. Phys. Eng. Sci. 365(1851), 539–560 (2007) 7. Moreu, F., Spencer, B.F.: Framework for Consequence-based Management and Safety of Railroad Bridge Infrastructure Using Wireless Smart Sensors (WSS). Newmark Structural Engineering Laboratory. University of Illinois at Urbana-Champaign, Champaign (2015) 8. Yun, H., Nayeri, R., Tasbihgoo, F., Wahbeh, M., Caffrey, J., Wolfe, R., Nigbor, R., Masri, S.F., Abdel-Ghaffar, A., Sheng, L.H.: Monitoring the collision of a Cargo Ship with the Vincent Thomas Bridge. Struct. Control. Health Monit. 15(2), 183–206 (2008) 9. Coverley, P.T., Staszewski, W.J.: Impact damage location in composite structures using optimized sensor triangulation procedure. Smart Mater. Struct. 12(5), 795–803 (2003) 10. Sun, Z., Chang, C.C.: Structural damage assessment based on wavelet packet trans-form. J. Struct. Eng. 128(10), 1354–1361 (2002) 11. Taha, M.M.R., Noureldin, A., Lucero, J.L., Baca, T.J.: Wavelet transform for structural health monitoring: a compendium of uses and features. Struct. Health Monit. 5(3), 267–295 (2006) 12. Song, G., Olmi, C., Gu, H.: An overheight vehicle–bridge collision monitoring system using piezoelectric transducers. Smart Mater. Struct. 16(2), 462–468 (2007) 13. Sharma, H., Hurlebaus, S.: Overheight collision protection measures for bridges. In: Structures Congress 2012, pp. 790–797. ASCE (2012) 14. Kurt, E.G., Varma, A.H., Hong, S.: FEM Simulation for INDOT Temporary Concrete Anchored Barrier. Joint Transportation Research Program (2012) 15. Buth, C.E., Williams, W.F., Brackin, M.S., Lord, D., Geedipally, S.R., Abu-Odeh, A.Y.: Analysis of large truck collisions with bridge piers: phase 1. Report of guidelines for designing bridge piers and abutments for vehicle collisions. (No. FHWA/TX-10/9-4973-1), 2010 16. Consolazio, G.R., McVay, M.C., Cowan, D.R., Davidson, M.T., Getter, D.J.: Development of improved bridge design provisions for barge impact loading. (No. UF Project 00051117), 2008 17. Luperi, F.J., Pinto, F.: Structural behavior of barges in high-energy collisions against bridge piers. J. Bridg. Eng. 21(2), 04015049 (2016) 18. ANSYS Inc., ANSYS Autodyn User’s Manual. Cecil Township, PA, 2016 19. ASTM International: ASTM A572-15 Standard Specification for High-Strength Low-Alloy Columbium—Vanadium Structural Steel. (2015) 20. Jones, N.: Structural Impact. Cambridge University Press, Cambridge (1997) 21. Wu, X., Ghaboussi, J., Garrett, J.H.: Use of neural networks in detection of structural damage. Br. J. Surg. 81(11), 578–581 (2010) 22. Ondra, V., Sever, I.A., Schwingshackl, C.W.: A method for detection and characterization of structural non-linearities using the Hilbert transform and neural networks. Mech. Syst. Signal Process. 83(2017), 210–227 (2016)

Chapter 29

High-Frequency Impedance Measurements for Microsecond State Detection Ryan A. Kettle, Jacob C. Dodson, and Steven R. Anton

Abstract This work investigates the use of the electromechanical impedance (EMI) method at high frequencies for application in state detection in highly dynamic systems; a key aspect of which is the excitation and measurement of PZT impedance at higher frequency bands than those typically used in structural health monitoring, in ranges greater than 1 MHz. The use of impedance analyzers, which are typically used in EMI measurements, are not considered for this work due to their slow measurement speeds, large size, heavy weight, and high cost. Instead, an alternative impedance measurement approach from the literature will be leveraged. The alternative method, which allows impedance measurements to be made using standard data acquisition devices, was originally developed to present a low-cost solution to impedance measurement. The focus of this work, however, is the adaptation of the alternative measurement approach to allow accurate impedance measurements at frequency ranges in excess of 1 MHz. The accuracy of the alternative system’s impedance measurements will be verified by comparison with theoretical models. First, the impedance response of a simple RLC circuit with well understood dynamics will be experimentally tested and then the resulting measurements will be compared to analytical solutions of the circuit impedance to validate the measurement system. This alternative measurement system is then used to perform high frequency impedance measurements on a piezoelectric wafer embedded in a structure in order to detect structural damage. The long-term goal of this research is the deployment of an impedance measurement system using highspeed real-time hardware, such as field-programmable gate array technology which has significantly higher sampling rates and calculation speeds compared to traditional data acquisition systems, to enable microsecond state detection of structures operating in highly dynamic environments using the EMI method. Microsecond state detection has a number of potential applications, such as in aerospace technology, the mining industry, and civil architecture. Keywords Structural health monitoring (SHM) • Electromechanical impedance (EMI) method • Piezoelectric materials • Microsecond state detection • Dynamic systems

29.1 Introduction The goal of this research is to develop a system capable of microsecond state detection in highly dynamic environments. In this work, a structure’s state refers to any change to its physical properties that affects its dynamic response which includes mass, stiffness, boundary conditions, etc. Microsecond state detection has a staggeringly wide array of potential applications. This includes the monitoring of drill strings used in mining, monitoring fluid flow around hypersonic aircraft for better flight handling, deployment in civil structures for rapid structural assessment and blast mitigation, and monitoring critical structures that are part of complex assemblies such as those commonly found in the automotive and aerospace industries. This work directly builds off of the foundation of knowledge created by researchers in the field of structural health monitoring. Structural health monitoring can be described as a range of techniques and tools to detect the development of damage in structures. In particular, the structural health monitoring technique known as the electromechanical impedance method is used in this work for state detection. The electromechanical impedance method is an active sensing method, meaning the structure is actively “interrogated” instead of relying on passive excitation [1]. This effect is achieved by the electromechanical coupling exhibited by piezoelectric (PZT) materials, which allows them to act as both a sensor and an actuator simultaneously. The electromechanical coupling also means that the electrical impedance of a piezoelectric R.A. Kettle () • S.R. Anton Department of Mechanical Engineering, Tennessee Technological University, Cookeville, TN, 38505, USA e-mail: [email protected] J.C. Dodson Air Force Research Laboratory (AFRL/RWMF), Eglin AFB, FL, 32542-5430, USA © The Society for Experimental Mechanics, Inc. 2017 J. Caicedo, S. Pakzad (eds.), Dynamics of Civil Structures, Volume 2, Conference Proceedings of the Society for Experimental Mechanics Series, DOI 10.1007/978-3-319-54777-0_29

235

236

R.A. Kettle et al.

Fig. 29.1 The electromechanical impedance method being used to detect damage in a structure

material bonded to a structure becomes a function of the mechanical impedance of the host structure. If the properties of the piezoelectric material are assumed to remain constant, then any changes in the in the electrical impedance of the piezoelectric material can be attributed to a change in the mechanical impedance of the structure [2]. To perform state detection, the electrical impedance signature of the structure in a known, healthy state is used as a baseline. As the structure is being monitored, the measured data is continuously compared to the baseline and any significant deviation between the two is attributed to a state change in the structure. The process of using the electromechanical impedance to detect damage in a structure is illustrated in Fig. 29.1. To achieve state detection at the microsecond timescale, high frequency excitation signal are required, on the order of 1 MHz or higher [3], which is a significant increase over the 30–400 kHz normally used [1]. A further break from the norm that is required for microsecond state detection is a change in the hardware utilized to make the electrical impedance measurement. Normally, an impedance analyzer, such as an HP4194A, is used to make impedance measurements. However, these devices have several drawbacks, which include being large (1600  1400  2400 ), heavy (81.4 lbs), expensive ($40,000), and the fact that they use a slow stepped sinewave method of measurement. These issues have been noted by previous researchers, with the biggest concern being cost. Because of this there have been a multitude of attempts to create new, low-cost impedance measurement devices which include the use of an FFT analyzer [4], a digital signal processor (DSP) [5], an AD5933 impedance analyzer IC chip [6, 7], and a data acquisition device (DAQ) [8, 9]. The method employing the FFT analyzer utilizes a measurement circuit to calculate an approximation of the unknown impedance. This measurement is based on the value of a known resistance placed in the circuit with the assumption that this resistor does not make a significant impact on the impedance measurement. Additionally, this method tends to be inaccurate at high frequencies due to a drop in resistance across the PZT at high frequencies due to the capacitive nature of PZTs. The usefulness of the DSP method is hindered by the small memory space of the DSP, which restricts its usable range and frequency resolution [9]. The AD5933, although a low cost solution, faces two distinct disadvantages: it must be calibrated before use [10] and it is limited to a maximum frequency of 100 kHz [9, 10]. The DAQ method however appears advantageous because it allows for a large range of hardware options, including the use of DAQs with exceptionally high sampling rates. This makes it a very appealing choice for high frequency applications, such as microsecond state detection. The excitation signal applied to the PZT is also a critical factor. The stepped sign measurement approach utilized by impedance analyzers is quite time consuming and thus another method for excitation is required to achieve microsecond state detection. A chirp signal is a logical choice as it can excite a very broad range of frequencies in a very short amount of time. Furthermore, the chirp signal is a common waveform that can be produced by many function generators and as such its use has the advantage of not requiring the creation of any custom voltage signals. Chirp signals have been used to make accurate EMI measurements by other researchers employing various methods, all of which showed that the chirp signal gave comparable results to impedance analyzers [4, 8, 11, 12]. This pervasive use of the chirp signal was also explored further by Baptista et al. who conducted a study comparing uniform white noise, periodic random noise, and a chirp signal for use as an excitation signal and concluded that the chirp signal performed the best measurement [13].

29 High-Frequency Impedance Measurements for Microsecond State Detection

237

29.2 Experimental Setup The experimental setup and measurement technique used in this work are adaptations from the efforts of other researchers who created “low-cost circuits” for impedance measurements [9, 14]. The low-cost circuit in its simplest form is just an excitation source in series with a known resistance, Rm , and an unknown impedance, Zunk . If the voltage of the excitation signal, Vx is known and the voltage response of the node between Rm and Zunk , Vy , is known, then Zunk can be solved for using Eq. (29.1). Zunk .f / D

Vy .f / Rm Vx .f /  Vy .f /

(29.1)

However, in order to measure Vx and Vy a data acquisition device needs to be used at both locations. The data acquisition device has its own internal equivalent circuit, represented by CDAQ and RDAQ in parallel, that then gets added to the measurement circuit, which is shown in Fig. 29.2. The excitation signal used in this study was a linear chirp produced by an Agilent 33220A Function Generator and the data acquisition device used was an NI PXIe-5122 Oscilloscope Card, the impedance for both is given in Table 29.1. The function generator and DAQ were connected to each other at synchronization ports via a BNC cable for triggering purposes. Additionally, they were both connected to the circuit using BNC to grabber cables that were approximately 6 in. long. The wire length was intentionally kept as short as possible to reduce the effects of wire resistance and to avoid introducing any additional environmental noise. The NI PXIe-5122 was connected to an NI PXIe-1082 chassis that also housed an NI PXIe-8133 controller with LabVIEW installed. The voltages Vx and Vy were recorded via a LabVIEW program on the controller and the date was saved as a ‘.txt’ file. This file was then loaded into a MATLAB program that performed all data processing, which includes impedance calculation, averaging, plotting, and error calculation. An overall view of the experimental setup can be seen in Fig. 29.3. More information regarding specific experiments is given in the results section immediately following. Finally, it should be noted that the goal of this work is to extend the measurement frequency range of the EMI method to the MHz range by utilizing the “low-cost” architecture presented in the literature, however, the hardware utilized in this work is not low-cost.

Fig. 29.2 “Low-cost” impedance measurement circuit Table 29.1 Impedance settings applied to the systems utilized for signal generation and acquisition

System Agilent function generator PXIe-5122 scope card

Impedance “High Z load” 1 M

238

R.A. Kettle et al.

Fig. 29.3 Experimental setup used for impedance measurement, including a structure to be measured

25

Re(Z) (Ohms)

20 15 10

R = 995 Ω m

R = 11.4 Ω

5 0

m

Theoretical

1

2

3

4

Frequency (MHz) Fig. 29.4 Impedance measurement of an 11  resistor using different values of Rm . 2 million samples were taken at 10 MS/s. The chirp excitation signal had an initial frequency of 100 Hz, a final frequency of 10 MHz, an 18 Vpp amplitude, and a sweep time of 0.1 s

29.3 Verification of Measurement Methodology Before measurements to perform state detection were attempted, the accuracy of the measurement method was verified. To do this, simple circuits with well understood impedance responses across wide frequency bands, such as resistors and resistorinductor-capacitor parallel circuits, were tested. One phenomenon that was observed in early measurements on a simple resistive circuit is that there was a low-pass filtering effect occurring on the signal. At higher frequencies the calculated impedance would suddenly and drastically decrease from the correct value. It was eventually discovered that this was caused by the internal capacitance of the data acquisition device, CDAQ , and the known resistance, Rm , effectively creating a low pass filter. This effect can be mitigated by choosing appropriately small values of Rm , on the order of 10  or less. The effect of the value of Rm on the impedance measurement can be seen in Fig. 29.4, where an 11  resistor was measured with a relatively low Rm of 11.4  and then with a higher Rm of 995 . Since a resistor has negligible values of capacitance and inductance, its theoretical impedance is simply a constant resistance (purely real impedance) across the entire frequency range. Figure 29.4 shows that the higher value of Rm produces inaccurate impedance measurements that vary significantly from the theoretical result. On the contrary, the lower value of Rm was in excellent agreement with the theoretical model below 4 MHz. A parallel resistor-inductor-capacitor (RLC) circuit was also measured to ensure accurate impedance measurements were being produced when measuring a dynamic circuit. A parallel RLC circuit was chosen because, assuming ideality, it produces a single impedance peak at a known frequency. The frequency of the peak is referred to as the resonance frequency of the circuit. This single resonance peak is a similar, yet simpler, frequency response compared to a PZT, because a PZT typically exhibits multiple peaks in a given frequency band. The resonance frequency of an RLC parallel circuit is given as

29 High-Frequency Impedance Measurements for Microsecond State Detection

239

12

Re(Z) (Ohms)

M easured Theoretical

= 10.0 Ω = 36.6 μH = 0.22 μF

10 8 6 4 2 0 -2 50

100

150

200

250

300

Frequency (kHz) Fig. 29.5 Impedance measurement of a parallel RLC circuit with an Rm of 10.2 . 2 million samples were taken at 2 MS/s. The chirp excitation signal had an initial frequency of 100 Hz, a final frequency of 1 MHz, a 10 Vpp amplitude, and a sweep time of 1 s

f0 D

1 p

2 LC

;

(29.2)

where f0 is the resonance frequency, L is inductance, and C is capacitance. Furthermore, the total impedance response of the circuit is given as 1  ZT D ZR1 C ZC1 C ZL1 ;

(29.3)

where ZT is the total impedance response, ZR is the impedance of the resistor, ZC is the impedance of the capacitor, and ZL is the impedance of the inductor. The impedance values for the individual components are given as ZR D R

(29.4)

ZC D .j2fC/1

(29.5)

ZL D j2fL;

(29.6)

where R is the resistance, f is excitation frequency, and j is the imaginary unit. Figure 29.5 shows the real part of the impedance (resistance) measurement of a parallel RLC circuit with the values of R D 10.0 , L D 36.6 H, and C D 0.22 F. The real part of the impedance is used exclusively in this work because it has been demonstrated to have the greatest sensitivity to damage [1]. This measurement is compared to the theoretical impedance value calculated using Eq. (29.3), and the measurement shows a good match to the analytical model. Furthermore, for the given values of R, L, and C, Eq. (29.2) gives a calculated resonance frequency f0 56 kHz, and the measurement matches this value reasonably well. The good agreement between the measurement and analytical model of the impedance of the parallel RLC circuit makes a compelling case for the accuracy of the employed measurement method in its ability to measure dynamic impedance response, such as that expected when measuring PZTs.

29.4 Damage Detection at High Frequencies Having proved the accuracy of the impedance measurement methodology using simple circuits, experimentation was then performed on a piezoelectric transducer in the MHz frequency range to investigate the feasibility of microsecond state detection. A small circular PZT wafer with a diameter, d, of 6.35 mm and a thickness, t, of 0.254 mm made from APC 850 material (PZT-5A) was embedded in a 3D printed structure. The structure was then clamped at one end, effectively forming a cantilever beam simply for the convenience of some initial non-destructive tests. The structure remained in this

240

R.A. Kettle et al.

cantilever position for consistency with previous tests but this is otherwise unnecessary. Twenty separate measurements were run on the structure, 10 in an undamaged state and 10 in a damaged state. The damage induced was a 0.300 notch cut into the structure 0.7500 away from the piezoelectric material wafer. Both states of the 3D printed beam can be seen in Fig. 29.6. The 10 measurements in each state were averaged and then plotted for comparison. Initial comparisons made by plotting the impedance spectrums of both the damaged and undamaged states showed very significant divergence in multiple frequency bands at lower frequencies. This is a good example of why structural health monitoring is typically done in the tens to hundreds of kHz. This is however a well-known phenomenon and not important to the goal of microsecond state detection. Results in the MHz range are necessary to prove the feasibility of microsecond state detection, and for this reason will be the sole topic presented and discussed in this work. In this range of frequencies in excess of 1 MHz there was an exceptionally high impedance peak located at roughly 9 MHz; the measurements for both damage states in the region of this peak are shown in Fig. 29.7. Because of the exceptionally high impedance value of this peak in comparison to others in the MHz range, it stands to reason that this peak is caused by the first thickness resonance mode, fT , of the piezoelectric transducer. This can be verified by using the thickness mode frequency constant, NT , to calculate the resonance frequency, given by fT D

NT ; t

(29.7)

where t is the thickness of the PZT. With the value of NT given as 2040 Hz.m from the manufacturer specifications [15], Eq. (29.7) gives fT D 8.03 MHz. The deviation in frequency between this calculation and the measured value can be accounted for by the following reasons. First, the geometry of the PZT wasn’t measured before the embedding procedure, and as such the nominal thickness of the ordered part was used in the calculation. The given tolerance of t is ˙0.025 mm and so a simple difference in geometry can account for values of fT up to 8.9 MHz. Second, since the PZT is embedded in a structure

Fig. 29.6 (a) The 3D printed structure with embedded PZT, undamaged. (b) The 3D printed structure in a damaged state 200 Damaged Undamaged

Re(Z) (Ohms)

150 100 50 0 8.8

9.0

9.2

9.4

9.6

9.8

10.0

Frequency (MHz) Fig. 29.7 Comparison between the average of 10 tests of the undamaged state the average of 10 tests of the damaged state with an Rm of 10.7 . 3 million samples were taken at 50 MS/s. The chirp excitation signal had an initial frequency of 200 kHz, a final frequency of 10 MHz, an 18 Vpp amplitude, and a sweep time of 0.06 s

29 High-Frequency Impedance Measurements for Microsecond State Detection Table 29.2 Root-mean-squared error of damaged and undamaged states compared to 7 measurement averaged baseline

241 State Undamaged

Average undamaged Damaged

Average damaged

Measurement # 8 9 10 8, 9, & 10 1 2 3 4 5 6 7 8 9 10 1–10

RMSE 0.901 0.896 0.898 0.898 3.584 3.535 3.512 3.465 3.487 3.493 3.503 3.487 3.487 3.446 3.500

it has boundary conditions that no longer reflect a free-free condition, but has instead an elastically constrained condition. This stiffer boundary condition would cause the first thickness resonance mode to increase. Third, the value of ft used in the calculation came from the manufacturer and was not derived from the material properties of the specific sample used and is consequently not exact. It is therefore reasonable to assume that the large peak at 9 MHz is in fact the first thickness resonance mode. Of particular importance is the fact that the impedance peak illustrates a slight divergence between the damaged and undamaged states. This supports previous work by the authors that predicted thickness modes would be important for damage detection at the high frequencies required for microsecond state detection [16]. While a qualitative evaluation of noticeable difference in signal is important for identifying frequency ranges sensitive to damage, a quantitative evaluation is necessary for any method of automated damage detection. To perform a quantitative comparison between the two states in this frequency range, an averaged baseline signal was constructed using the first 7 measurements of the undamaged state. This baseline was then compared to the remaining 3 undamaged states and the 10 damaged states using the root-mean-squared error (RMSE) given by s RMSE D

Pn iD1

.x1;i  x2;i /2 ; n

(29.8)

where i is the index of a given point in a series of data, n is the total number of points in the series, x1 , i represents the value of the first series at index i, and x2 , i represents the value of the second series at index i. The results of the RMSE calculations are shown in Table 29.2 and it can be seen that the undamaged cases vary from the baseline by an RMSE of less than 1% and the damaged cases by over 3%. Although the difference between these cases is relatively small, this establishes the fact that there is a measurable, quantitative difference between the signals measured in the damaged and undamaged states, detected at frequencies in the MHz frequency range. The significance of this finding is that it demonstrates the feasibility of detecting structural changes at frequencies in excess of 1 MHz, which is a crucial aspect of microsecond state detection in highly dynamic environments.

29.5 Conclusions The goal of this work is to develop the technologies and methods required to enable microsecond state detection of structures operating in highly dynamic environments using the electrometrical impedance method. A vital step in achieving this goal is successfully detecting structural changes at frequencies greater than 1 MHz. To accomplish this goal, a non-traditional impedance measurement system was utilized instead of a traditional impedance analyzer. To verify the accuracy of the new measurement hardware and methodology, measurements of simple circuits were compared to analytical solutions. First, a high frequency measurement of a resistor was performed, which showed excellent agreement with the analytical solution up to 4 MHz. Second, an impedance peak produced by a parallel resistor-inductor-capacitor circuit was measured that also showed good agreement with the analytical solution. These tests demonstrate that accurate impedance measurements can be

242

R.A. Kettle et al.

made with non-traditional hardware using the method presented as an alternative to an impedance analyzer. A structure with an embedded PZT was then presented and its impedance measured at high frequencies in a damaged and undamaged state. The two signal were compared and an impedance peak around 9 MHz, caused by the first thickness resonance mode of the transducer, was observed which exhibited a small difference between the two signals. A baseline measurement was then constructed using several measurements from the undamaged state and were compared to individual measurements of both states using the root-mean-squared error. The-root-mean squared error for the tests in the undamaged state were all under 1%, and greater than 3% for the tests taken in the damaged state. It can be concluded from these results that damage can be detected at high frequencies. The accurate measurement of impedance at high frequencies using alternative hardware in place of an impedance analyzer as well as its successful use for damage detection at high frequencies are both important milestones and represent significant steps forward in the pursuit of achieving microsecond state detection in highly dynamic environments.

29.6 Future Work There are three additional routes of future work that remain. First, comparisons between measurements of piezoelectric wafers to theoretical models need to be made. Second is the development of a method to decouple the data acquisition device from the measurement circuit. Currently, the low value of Rm only mitigates the problem caused by the internal data acquisition device’s internal impedance. In order to enable accurate impedance measurements at even higher frequencies the development of a method to completely decouple the data acquisition device’s internal capacitance is necessary. Lastly, a piezoelectric material which exhibits a higher thickness mode response, which is the phenomenon that creates the amplitude peaks at higher frequencies, should be used to potentially obtain a larger value in the RMSE difference between states.

References 1. Park, G., Sohn, H., Farrar, C.R., Inman, D.J.: Overview of piezoelectric impedance-based health monitoring and path forward. Shock Vib. Dig. 35(6), 451–463 (2003) 2. Bhalla, S., Moharana, S.: A refined shear lag model for adhesively bonded piezo-impedance transducers. J. Intell. Mater. Syst. Struct. 24(1), 33–48 (2013) 3. Kettle, R.A., Dick, A.J., Dodson, J.C., Foley, J.R., Anton, S.R.: Real-time state detection in highly dynamic systems. Presented at the Proceedings of the 34th IMAC, Cham, 2016 4. Peairs, D.M., Park, G., Inman, D.J.: Improving accessibility of the impedance-based structural health monitoring method. J. Intell. Mater. Syst. Struct. 15(2), 129–139 (2004) 5. Kim, J., Grisso, B.L., Ha, D.S., Inman, D.J.: A system-on-board approach for impedance-based structural health monitoring. Presented at the The 14th International Symposium on: Smart Structures and Materials & Nondestructive Evaluation and Health Monitoring, 2007 6. David, L.M., Michael, D.T., Gyuhae, P., Charles, R.F.: Development of an impedance-based wireless sensor node for structural health monitoring. Smart Mater. Struct. 16(6), 2137 (2007) 7. Park, S., Lee, J.-J., Yun, C.-B., Inman, D.J.: Electro-mechanical impedance-based wireless structural health monitoring using PCA-data compression and k-means clustering algorithms. J. Intell. Mater. Syst. Struct. 19(4), 509–520 (2008) 8. Xu, B., Giurgiutiu, V.: A low-cost and field portable electromechanical (E/M) impedance analyzer for active structural health monitoring. Presented at the Proceedings of the 5th International Workshop on Structural Health Monitoring, Stanford University, 2005 9. Baptista, F.G.: A new impedance measurement system for PZT-based structural health monitoring. IEEE Trans. Instrum. Meas. 58(10), 3602– 3608 (2009) 10. Wandowski, T., Malinowski, P., Ostachowicz, W.: Calibration problem of AD5933 device for electromechanical impedance measurements. Presented at the EWSHM-7th European Workshop on Structural Health Monitoring, 2014 11. Baptista, F.G., Vieira Filho, J., Inman, D.J.: Real-time multi-sensors measurement system with temperature effects compensation for impedance-based structural health monitoring. Struct. Health Monit. 11(2), 173–186 (2012) 12. Saar, T.: Robust piezo impedance magnitude measurement method. Elektronika ir Elektrotechnika. 113(7), 107–110 (2011) 13. Baptista, F.G., Filho, J.V., Inman, D.J.: Influence of excitation signal on impedance-based structural health monitoring. J. Intell. Mater. Syst. Struct. 21(14), 1409–1416 (2010) 14. Lewis Jr., G.K., Lewis Sr., G.K., Olbricht, W.: Cost-effective broad-band electrical impedance spectroscopy measurement circuit and signal analysis for piezo-materials and ultrasound transducers. Meas. Sci. Technol. 19(10), 105102 (2008) 15. (2016, 10/24/16). Physical and Piezoelectric Properties of APC Materials. Available: https://www.americanpiezo.com/apcmaterials/piezoelectric-properties.html 16. Kettle, R.A., Anton, S.R.: Rapid evaluation of mechanical boundary conditions using impedance based structural health monitoring. Presented at the SPIE Smart Structures and MaterialsC Nondestructive Evaluation and Health Monitoring, 2016

Chapter 30

Structural Stiffness Identification of Skewed Slab Bridges with Limited Information for Load Rating Purpose Abdollah Bagheri, Mohamad Alipour, Salman Usmani, Osman E. Ozbulut, and Devin K. Harris

Abstract This paper presents a method for identifying structural stiffness of skewed reinforced concrete slab bridges with limited structural information using measured acceleration data. This information might be used for nondestructive evaluation, condition assessment, and load rating of bridges. A large number of slab bridges with different structural dimensions such as skew angle, span, width, and thickness was first analyzed using finite element method to obtain their first modal frequency. This population of data was then used to create an artificial neural network, which can predict a coefficient that plays an important role in identifying the flexural rigidity of slab bridges. This approach was applied to estimate the flexural rigidity of a highly skewed reinforced concrete slab bridge in the state of Virginia for load rating purpose. The bridge was instrumented with wireless accelerometers, and the vibration responses of the bridge under ambient loading and impact hammer test were recorded. An algorithm based on the variational mode decomposition was employed to identify modal properties of the bridge. Then, the flexural rigidity of bridge was computed from the established relationship between the first natural frequency and the flexural rigidity of bridge. Results show that the proposed method is capable of predicting structural stiffness, and can be used for load rating of bridges without structural information. Keywords Structural stiffness identification • Modal frequency • Vibration response • Skewed slab bridge • Variational mode decomposition

30.1 Introduction Load rating is the process of determining the safe load-carrying capacity of a bridge and, thus, serves as a basis for prioritizing maintenance operations and allocation of resources [1–3]. Typically, load ratings are developed in accordance with the rules of structural mechanics using design drawings and details that define the geometry and material properties of the bridge. Therefore, the information needed to carry out this load rating includes the latest safety inspection report, prior load rating files, and design plans or as-built drawings of the structure. However, there are cases where these design plans are missing or incomplete due to lack of documentation at the time of construction, improper storage or the evolution of data management practices. For these structures, the Manual for Bridge Evaluation (MBE) provides limited guidance on the process for load rating [4]. The language in MBE indicates that an inspection by a qualified inspector and evaluation by a qualified engineer may be sufficient to establish an approximate load rating. This guidance does not explicitly state, but does imply, that engineering judgment may be necessary in which an experienced engineer considers relevant factors, such as the original design live loads, the past performance and current physical condition of the bridge, current loads, and age, to arrive at a judgment-based load rating. Since the State Departments of Transportation in the United States has responsibility for load rating a large number of bridges and culverts within their inventory [5], structures without existing plans or insufficient details represent a significant bottleneck in the management and decision-making. Bridge engineers might elect to be overly conservative about a given bridge’s capacity. As a result, resources may unnecessarily be devoted to the structure instead of others that might truly be in need. Furthermore, overly conservative ratings may restrict the types of vehicles that traverse the bridge, and as a result, hinder commerce within the United States. On the other hand, these judgment-based ratings can be subjective and may pose a risk, as this approach may not accurately describe a bridge’s behavior [6]. Therefore, load rating strategies for bridges with limited or missing as-built information are needed, and our aim is to presents methods for load rating these types of structures.

A. Bagheri • M. Alipour • S. Usmani • O.E. Ozbulut () • D.K. Harris Department of Civil and Environmental Engineering, University of Virginia, Charlottesville, VA 22904, USA e-mail: [email protected] © The Society for Experimental Mechanics, Inc. 2017 J. Caicedo, S. Pakzad (eds.), Dynamics of Civil Structures, Volume 2, Conference Proceedings of the Society for Experimental Mechanics Series, DOI 10.1007/978-3-319-54777-0_30

243

244

A. Bagheri et al.

Load rating factor is a function of the bridge’s capacity and the effect of dead and live loads in bridge, and both of these parameters depend on the stiffness of bridge. This paper describes a method for identifying the bending stiffness of skewed reinforced concrete slab bridges using the measured first modal frequency of the bridge. This information can be used in load rating of bridges with limited or missing as-built information. Vibration testing of an in-service was conducted and the developed method was then employed to estimate the bending stiffness of the bridge.

30.2 Stiffness Identification Method 30.2.1 Problem Formulation For a structural system with a single degree of freedom, the relation between the angular natural frequency ! of the structure and the structural stiffness K can be written as: !2 D

K M

(30.1)

where M is the mass of the system. The analogous relationship can be presented for a rectangular plate [7]: ! 2 D 2

D 

(30.2)

where  is the mass per unit area of the plate,  is a coefficient which depends to the plate’s dimension and boundary condition, and D represents the bending stiffness of the plate and is given as: DD

Et3 12 .1   2 /

(30.3)

where t is the plate’s thickness, and E and  are the elastic modulus and Poisson’s ratio of the plate’s material. Equation (30.2) can be solved to obtain the bending stiffness D of a plate. By measuring the vibration response of the plate and processing the measured signal, one can identify the modal frequency ! of the plate. Then, by knowing the plate’s material, the mass per unit area of the plate can be obtained. The coefficient  can be calculated from a function that depends to the plate’s dimension and boundary condition. This function is available in the literature for rectangular plates with different boundary conditions. Finally, the bending stiffness D is obtained by plugging three parameters namely !, , and  in Eq. (30.2) and solving it for the bending stiffness D. A similar method can be employed to identify the bending stiffness of a skewed slab bridge shown in Fig. 30.1. In this figure, a and b represent the bridge’s span and width, respectively, and the skew angle is shown by . Most slab bridges are built with simple supports at two ends of the span and parapets at two edges along the span, which has a significant contribution in the bridge’s stiffness. However, no equation is available in the literature to calculate the coefficient  for such a configuration. Here, a parametric study was conducted to obtain  for simply-supported skewed slab bridges with parapets. Fig. 30.1 Top view of a skewed slab bridge with its dimension

Parapet

b

Simple Support

a

30 Structural Stiffness Identification of Skewed Slab Bridges with Limited Information for Load Rating Purpose

245

30.2.2 Parametric Study In the parametric study, it was assumed that the bridge’s parapet has the same rectangular cross-section for the bridges with different span width and length. This is a valid assumption as the parapet’s dimensions for most slab bridges are standard. Therefore,  is a function of the bridge’s span a, width b, skew angle , and thickness t. For given parameters, the first modal frequency of the bridge can be obtained using a modal finite element analysis. Then, (a,b,,t) is calculated as follows: r  .a; b; ; t/ D ! .a; b; ; t/

 D

(30.4)

The modal finite element analysis was conducted for different values of the bridge’s span a, width b, skew angle , and thickness t for obtaining (a,b,,t). For each parameter, the following ranges were considered: a.m/ D Œ4 5 6 7 8 9 10 11 12 13 14 15

(30.5)

b.m/ D Œ4 5 6 7 8 9 10 11 12 13 14 15

(30.6)

t.m/ D Œ0:30 0:35 0:40 0:45 0:50 0:55 0:60 0:65 0:70

(30.7)



ı 

D Œ0 5 10 15 20 25 30 35 40 45 50

(30.8)

Based on these ranges for each parameter, the number of finite element models that should be run is 12  12  9  11 D 14,256. To minimize the computational efforts, a finite element code was written in MATLAB to model and analyze of slab bridges with all combination of each parameter automatically and continuously. Figure 30.2a shows the value of  as a function of the bridge’s span a and width b for the plate’s thickness of 0.3 m and the skew angle of 0ı , whereas Fig. 30.2b illustrates the same results for the skew angle of 45ı . It can be seen that at short spans and widths the value of  is larger than that for long spans and widths. This is due to the increase in the natural frequency of bridge with the decreasing the bridge’s span and width. The skew angle in a bridge causes to increase the stiffness of the

Fig. 30.2 The value of  as a function of the bridge’s span a and width b for the plate’s thickness of 0.3 m and: (a) the skew angle of 0ı and (b) the skew angle of 45ı

246

A. Bagheri et al.

Fig. 30.3 The value of  as a function of the bridge’s span a and width b for the plate’s thickness of 0.6 m and: (a) the skew angle of 0ı and (b) the skew angle of 45ı

structure which causes to increase the bridge’s modal frequency, and this effect increases the value of  which can be seen by comparing Fig. 30.2a, b. In order to see the effect of the plate’s thickness on , Fig. 30.3 shows the results of  for the plate’s thickness of 0.6 m. By comparing Figs. 30.2 and 30.3, it can be found that the value of  decreases by increasing the plate’s thickness at short spans and widths. The changing plate’s thickness affects both the mass and stiffness of bridge, and also changes the contribution of the parapet’s stiffness on the global stiffness of the structure. Therefore, it can be concluded that there is a complex relation between  and the plate’s thickness.

30.2.3 Mapping Process An artificial neural network (ANN) was employed to predict the relationship between the coefficient of  and the bridge’s parameters, namely the bridge’s span a, width b, skew angle , and thickness t. Neural networks are one of the most effective soft computing algorithms for data fitting and classification. They mimic a human brain by implementing the interconnection of artificial neurons. In this study, a feed-forward back-propagation ANN with three layers was used. The input layer received the data vector containing the four parameters of bridge. The hidden layer processed the data by multiplying the input vectors by weights and adding biases. The results constituted the argument of a transfer function that squashed the output values into a certain range. For the hidden layer, ten neurons were used by means of a trial and error method, and the hyperbolic tangent sigmoid transfer function was employed. For the input and output layers, a liner transfer function was used. The output layer had one node that provided the value of . To train the network, the Levenberg-Marquardt algorithm was used because of its high performance and speed [8]. To study the ability of the network to estimate the value of , we used all 14,256 data for the training. Figure 30.4 shows the ANN estimated  versus the actual , and it can be seen that there is no difference between them, because there is a linear function between the estimated and target values. This trained ANN will be used to estimate  in identifying process of the structural stiffness of a tested slab bridge.

30 Structural Stiffness Identification of Skewed Slab Bridges with Limited Information for Load Rating Purpose

247

Fig. 30.4 The ANN estimated  versus the actual 

30.3 Experimental Study 30.3.1 Test Description The structure tested in this study was the War Branch Bridge located in the Staunton District in the state of Virginia. The superstructure is comprised of two 9.75-m long, simply-supported reinforced concrete slabs that are 0.53 m thick and have a 45ı skew angle (see Fig. 30.5). The deck has 0.3 m diameter voids oriented in the direction of traffic and spaced 0.45 m apart. Built in 1976, the 2014 inspection report described the bridge to be in “fair” condition, with a deck/superstructure condition rating of 7. The bridge was selected from amongst the Virginia Department of Transportation (VDOT) population of reinforced concrete slab bridges with plans, with special consideration given to geometry similarity to the population of this major category of bridges without plans. For the vibration testing experiments, one of the two spans in the bridge was instrumented with accelerometers. All instrumentation and acquisition comprised of Bridge Diagnostics, Inc. (BDI) equipment, where individual sensors physically connected to four-channel nodes, which in turn interfaced wirelessly with a base station/data acquisition unit. Vibration testing consisted of a series of experiments with excitation provided separately by ambient loading (wind and normal traffic), impact hammer, and electro-magnetic shaker.

30.3.2 Modal Data Identification In this paper, an algorithm based on the VMD is employed for identifying the modal properties of the bridge. In this method, the measured acceleration signals are decomposed into the modal responses by means of the VMD algorithm, and each obtained modal response has a center frequency which represents the natural frequency of the structure. Then, damping ratios are estimated by doing a fitting process on decaying amplitude of modal response. This method is capable of identifying all natural frequencies and damping ratios using only a single measurement of acceleration response at one suitable location. Mode shapes are then identified from the results of modal responses at all sensing location of the structure. The VMD is first used to decompose an acceleration signal S(t) into a set of sub-signals (modes), Sk (t), k D 1, 2, : : : , K, which have a compact bandwidth in spectral domain. It can be assumed each sub-signal to be compacted around a center vibration ! k which is determined using the decomposition algorithm. Therefore, each sub-signal represents the modal response of structure. In this process, the constrained variational problem is formed by means of Hilbert transform and frequency mixing based on minimizing the bandwidth of sub-signals. In order to render the problem unconstrained, the quadratic penalty and Lagrangian multipliers are employed. More information about the VMD algorithm can be found in [9].

248

A. Bagheri et al.

Fig. 30.5 Photo of the War Branch Bridge

By knowing the modal responses of a structure, it is easy to identify modal damping ratios for all vibration modes. In most engineering structures that have small damping ratio, damping ratio can be estimated from the slop of the decaying amplitude versus time t plot. In this paper, the plot of the decaying amplitude versus time t was fitted to a linear function using the linear last-square fit approach. The slop of the linear function was used to compute the damping ratio. Acceleration responses of the bridge in the vibration test using impact hammer were used to identify modal frequency and damping ratio of the bridge. The described approach was used to decompose each acceleration signal to sub-signals which represent modal responses of the bridge. The first and second modal frequency of the bridge was 26.07 and 30.59 Hz, respectively. For these modes, the obtained damping ratios were 4.3 and 3.9%, which are close to the common value of the damping ratio 5% mentioned in design codes for concrete structures.

30.3.3 Stiffness Identification For identifying the bending stiffness of the tested bridge, the trained ANN is first used to obtain the value of  based on the parameters of the bridge. For the War Branch Bridge, the bridge’s span a, width b, skew angle , and thickness t are 9.44 m, 9.75 m, 45ı , and 0.53 m, respectively. The value of  which is the output of the trained neural network for these bridge’s parameters is equal to 0.25. Then, by replacing the obtained value of  and the first modal frequency of 26.07 Hz in Eq. (30.4), the bending stiffness D is obtained as 3.86  108 N m. The elastic modulus of the bridge’s material can be computed from Eq. (30.3) based on the identified bending stiffness, which is equal to 31.62 GPa. These results can used for estimating the ultimate compressive strength of the bridge’s concrete and the amount of reinforcement area inside the concrete to estimate the bending capacity for computing the load rating factor.

30.4 Conclusion In this paper, a method for identifying the structural stiffness of skewed concrete slab bridges using modal data was presented. A large number of finite element models for slab bridges with different geometric characteristics were analyzed to provide a set of data for mapping bridge’s characteristics to a coefficient that is used in identifying the bending stiffness from the modal frequency of a bridge. An ANN was trained to predict this coefficient for a given simply-supported slab bridge. The methodology was used to identify the bending stiffness of an in-service slab bridge. The developed ANN tool and methodology can provide a simple approach to identify the structural stiffness of slab bridges. In a future study, the obtained bending stiffness and elastic modulus of materials will be combined with the measured strain data to calculate the capacity for inventory and operating load rating of the tested bridge.

30 Structural Stiffness Identification of Skewed Slab Bridges with Limited Information for Load Rating Purpose

249

Acknowledgments This material is based upon the work supported by the Virginia Department of Transportation. The authors would like to thank Dr. Bernard L. Kassner of Virginia Transportation Research Council for his helps in conducting the vibration testing of the bridge.

References 1. Chajes, M., Shenton III, H., O’Shea, D.: Bridge-condition assessment and load rating using nondestructive evaluation methods. Transp. Res. Rec. J. Transp. Res. Board. 1696, 83–91 (2000) 2. Faber, M.H., Val, D.V., Stewart, M.G.: Proof load testing for bridge assessment and upgrading. Eng. Struct. 22(12), 1677–1689 (2000) 3. Cai, C.S., Shahawy, M.: Understanding capacity rating of bridges from load tests. Pract. Period. Struct. Des. Constr. 8(4), 209–216 (2003) 4. AASHTO: Manual for Bridge Evaluation, 2nd edn. with 2016 Interim Revisions. American Association of State Highway and Transportation Officials, Washington (2015) 5. VDOT: State of the Structures and Bridges Report. Virginia Department of Transportation, Richmond (2015) 6. Ma, Y., Wang, L., Zhang, J., Xiang, Y., Liu, Y.: Bridge remaining strength prediction integrated with Bayesian network and In situ load testing. J. Bridg. Eng. 19(10), 04014037 (2014) 7. Leissa, A.W.: Vibration of plates. National Aeronautics and Space Administration, NASA SP-160 (1969) 8. Bagheri, A., Rizzo, P., Al-Nazer, L.: Determination of the neutral temperature of slender beams by using nonlinear solitary waves. J. Eng. Mech. 141(6), 04014163 (2014) 9. Dragomiretskiy, K., Zosso, D.: Variational mode decomposition. IEEE Trans. Signal Process. 62(3), 531–544 (2014)

Chapter 31

Online Systems Parameters Identification for Structural Monitoring Using Algebraic Techniques L.G. Trujillo-Franco, G. Silva-Navarro, and F. Beltrán-Carbajal

Abstract Nowadays, with the modern techniques and developments on sensors and actuators technologies, disciplines like Operational Modal Analysis (OMA), Structural Health Monitoring (SHM) and Non-Destructive Evaluation (NDE), among others, are now basic parts of the study, modeling and monitoring for modern civil structures and vibrating mechanical systems. The most important system parameters of a given mechanical system, including civil structures, like modal parameters, mass and stiffness matrices are indicatives of the inherent nature and dynamical behavior of it, and, at the same time, a possible way to detect failures by comparing two different sets of such parameters, before and after any failure happens. In this work, a novel fast and online system parameter identification scheme, based on module theory and algebraic techniques for structural monitoring and vibration absorption purposes or model updating for mechanical systems under nominal operation conditions is proposed, that is, in an operational fashion, where only system output information is available. The proposed scheme is evaluated with experimental data. Keywords Algebraic identification • Operational modal analysis • Real time structure monitoring • Structural health monitoring • System identification

31.1 Introduction On the area of mechanical design and modeling (civil structures inclusive), the process called modal parameter identification or modal analysis has become into a basic technological tool, which allows having a correct knowledge and sometimes a prediction of the system response under harmonic excitation. In this information era, modal analysis can specially count on the newest and powerful data analyzers for storing and processing the data under study. In this order of ideas, the task of analyzing engineering structures response and behavior, when the system is working on its nominal operation conditions is a natural consequence [1, 2] (e.g., via the OMA approach). Most of the times, real world systems work under a completely random and hard to measure excitation; furthermore, in some special cases, it is desirable to monitor some interest output in a real-time fashion. Certainly, the actual state of the art for modal analysis has a robust background, consisting of plenty of identification algorithms in time or frequency domain, mainly for off-line estimation of modal parameters. However, it is important to consider that, most of these techniques are essentially asymptotic, recursive, complex and slow for on-line parameter estimation implementations, which could be required for efficient adaptive active noise and vibration control and monitoring applications on dynamic mechanical structures [3–5]. As reported in Refs. [4–7], here we apply a theoretical framework for the algebraic parameter identification on continuoustime linear systems. This identification approach platform consists of powerful mathematical tools as module theory, differential algebra and operational calculus. It is important to remark that, the operational calculus is a quite general approach based on different integral transformations of functions and generalized functions (e.g., Fourier, Laplace, Stieltjes, Hilbert, Bessel) [4–7]. The application of operational calculus in mechanical engineering is quite common for the transformation of functions from time to frequency domain and solving differential equations.

L.G. Trujillo-Franco () • G. Silva-Navarro Departamento de Ingeniería Eléctrica, Sección de Mecatrónica, Centro de Investigación y de Estudios Avanzados del I.P.N., Av. IPN No. 2508, Col. S.P. Zacatenco, C.P. 07360, Mexico, D.F., Mexico e-mail: [email protected]@cinvestav.mx F. Beltrán-Carbajal Departamento de Energía, Universidad Autónoma Metropolitana, Unidad Azcapotzalco, Av. San Pablo No. 180, Col. Reynosa Tamaulipas, C.P. 02200, Mexico, D.F., Mexico e-mail: [email protected] © The Society for Experimental Mechanics, Inc. 2017 J. Caicedo, S. Pakzad (eds.), Dynamics of Civil Structures, Volume 2, Conference Proceedings of the Society for Experimental Mechanics Series, DOI 10.1007/978-3-319-54777-0_31

251

252

L.G. Trujillo-Franco et al.

In this work we propose a novel system parameter monitoring scheme based on algebraic identification. We perform modal testing to the nominal or undamaged structure in order to have a reference for comparisons, and then, we induce changes directly on the structure, simulating failures (e.g., a loose or broken screw or a mass addition), in order to identify in a fast and online fashion the impact of each situation on the system modal parameters.

31.2 Illustrative Vibrating Mechanical System Consider the n Degrees-Of-Freedom (DOF) vibrating mechanical system consisting of a six story building-like structure as shown in Fig. 31.1, where xi , i D 1, 2, : : : , n, are the displacements of 6 masses representing the floors or Degrees-OfFreedom (DOF) of the structure, respectively. We model the columns as flexural springs with equivalent stiffness ki and consider the structural damping ratios ci as Rayleigh damping [8]. The simplified mathematical model of this flexible mechanical system of 6 DOF under harmonic and unknown excitation f is given by: MRx C CPx C Kx D f .t/;

x; f 2 R6

(31.1)

where x 2 R6 is the vector of generalized coordinates (displacements) of each floor respect to the main frame reference, and M, C and K are symmetric inertia, damping and stiffness 6  6 matrices, respectively, given by: 2

m1 6 0 6 6 0 6 6 MD 6 : 6 6 : 6 4 : 0

2 3 c1 Cc2 c2 0 0 : : :0 0 0 6 c c Cc c m2 0 : : : 0 0 7 2 2 3 3 6 7 6 0 m3 : : : 0 0 7 0 c3 c3 Cc4 6 7 6 7 : 7 ; CD 6 6 7 6 7 : 6 7 4 5 : 0 0 0 0 0 : : : 0 m6

: : : : : : : : :

0 0 0

0 0 0

3

2

k1 Ck2 k2 0 6 k k Ck k 7 2 2 3 3 6 7 6 7 0 k3 k3 Ck4 6 7 6 7 : 7 ; KD 6 6 7 6 c6 7 : 6 7 4 5 : : : : c1 c6 0 0 0

: : : : : : : : :

0 0 0

0 0 0

3

7 7 7 7 7 7 7 k6 7 7 5 : : : k6 k6 (31.2)

The modal analysis representation of the mathematical model (31.1) is defined in terms of the modal or principal coordinates qi , where i D 1, 2, : : : , 6 as follows (see, e.g., [1]) qR i C 2i !i qP i C !i2 qi D ‰ T f

Fig. 31.1 Six story Building-like structure. (a) Schematic diagram. (b) Physical plant

(31.3)

31 Online Systems Parameters Identification for Structural Monitoring Using Algebraic Techniques

253

with x.t/ D ‰q.t/

(31.4)

where ! i and  i denote the natural frequencies and damping ratios associated to the i-th vibration mode, respectively, and ‰ is the so-called 6  6 modal matrix given by 2 6 6 6 ‰ D6 6 4

11 21

:: :

51 61

: : : 22 : : : :: : : : : 52 : : : 62 : : : 12

3

15

16

25

26 7

55

7 :: 7 : 7 7 5 56

56

66

:: :

(31.5)

In notation of Mikusi´nski operational calculus [5, 6], this modal model is then described as 2  s C 2i !i s C !i2 qi .s/ D p0;i C p1;i s C

1i f1

C

2i f2

C C

6i f6

(31.6)

where po,i are constants depending on the system initial conditions at the time t0  0. From (31.3) and (31.5), one then obtains that xi .s/ D

n X jD1

ij

  p0;j C p1;j s

s2 C 2j !j s C !j2

(31.7)

Therefore, the physical displacements xi are given by pc .s/ xi .s/ D r0;i C r1;i s C C r2n2;i s2n2 C r2n1;i s2n1

(31.8)

pc .s/ D s2n C a2n1 s2n1 C C a1 s C a0

(31.9)

with

where pc (s) is the characteristic polynomial of the mechanical system and ri,j are constants which can be easily calculated by using the values of the system initial conditions as well as the modal matrix components ij . Naturally, the roots of the characteristic polynomial (31.9) provide the damping factors and damped natural frequencies, and hence the most descriptive information about the structure and its status.

31.3 Online Structural Monitoring The proposed online algebraic monitoring scheme shown in Fig. 31.2 works in conjunction with the building like structure shown in Fig. 31.1 for the illustrative cases to be considered in this work. The ARM (Advanced Risc Machine) takes samples of the acceleration of the sixth floor (in the horizontal axis direction) of the structure at a precisely fixed sample rate of 1Khz, and then, those samples are sent to a standard PC running under ® ® Windows 7 and Matlab /Simulink to finally perform the online identification scheme. We perform an on-line algebraic identification approach to estimate the modal parameters of the mechanical system through the real-time estimation of the coefficients ak of the system’s characteristic polynomial as reported in [5, 6] using only acceleration measurements of any floor of the structure. The application of the online algebraic identification scheme is performed using cumulative trapezoidal numerical integration with fixed sampling time of 1 ms. The algebraic identification scheme applied here is described on detail in [4] where is shown that by solving the algebraic equation (31.9) also detailed in [4–7] one obtains the parameter vector  as: 2

1 2 :: :

3

6 7 6 7 1 6 7 ™ D A1 B D 6 7 7 6 4 n1 5 n

(31.10)

254

L.G. Trujillo-Franco et al.

Fig. 31.2 Block diagram of the proposed identification and monitoring scheme

Then, the algebraic identifiers to estimate the coefficients ak of the characteristic polynomial without problems of singularities when the determinant  D det(A(t)) crosses by zero: R b ak D

j j R k1 ; jj

k D 1; 2; ; 2n  1

(31.11)

Thus, one could implement the algebraic identifiers (31.10) using only any available acceleration measurements xR i of any specific floor or degree of freedom. From the estimated coefficients b ak , one can obtain the roots of the characteristic polynomial: i D i C j!di ;

i D i  j!di ;

i D 1; 2; ; n

(31.12)

where i and ! di are estimates of the damping factors and damped natural frequencies of the mechanical system, respectively. Hence, the estimates of the natural frequencies and damping ratios are given by !ni D

q

i2 C !di2 ;

i i D  q

i2 C !di2

(31.13)

Remark: On the block diagram of Fig. 31.2 the shaded arrow indicates that once we identify the modal parameters or even the complete eigenstructure of the system, we can identify the excitation force and thus counteract it, as we have reported in previous works [5, 7, 9] for vibration absorption schemes or simply characterize this unknown excitation force.

31.4 Some Illustrative Cases In order to evaluate the performance of the proposed monitoring scheme we perform a traditional modal testing on the structure in order to know its natural status or nominal modal parameters. The results of the modal testing (impact hammer based) of the structure is reported in Table 31.1 and in Fig. 31.3. The algebraic identification of the system modal parameters is shown in Fig. 31.4, notice that the online identification of the modal parameters b  i and b ! ni takes less than 1 taking in to consideration that we compute the natural frequencies and the damping ratios using the estimates of the 12 coefficients ak corresponding to the characteristic polynomial of the mechanical system. Once calibrated the algebraic identifier, we consider our first case study: We induce one structural failure in the structure by losing one of the four screws on the third floor; next, we excite the structure with a random impact force and then, identify the new modal parameters of the “damaged” structure with only measurements of the sixth floor acceleration. The structural failure detail and the new system impact response are shown in Fig. 31.5.

31 Online Systems Parameters Identification for Structural Monitoring Using Algebraic Techniques Table 31.1 Modal parameters of the nominal system

255

Mode 1 2 3 4 5 6

Frequency (Hz) 1.148 3.39 5.44 7.16 8.53 9.34

Damping ratio (%) 0.15 0.39 0.18 0.19 0.18 0.17

5 5

4

Experimental FRF

3

4

⎢X6(ω)⎢[g]

X6(t) [g]

2 1 0

3

2

-1 -2

1

-3 -4

0

5

10

25

20

15

35

30

0

40

1

2

3

4

5

6

8

7

9

10

11

12

ω[Hz]

t[s]

0

0

2

4

2

4

0

6

5

0

0

4

2

0 0 x10-3 5

6

2

4

6

2 0

6

2

4

2

4

6

0 x10-3

2

4

6

0

2

4

6

0 -3 4 x10

2 0

6

3 2

1

∧

0

4

z4

wn1 [HZ ]

∧

wn4 [HZ ]

5

2

0.005

2 10

0

0.01

0

6

z3

z6 0

∧

∧

10

Estimated Actual value

z2

0

6

x10-3

z1

4

1

z5

2

4

2

5

∧

0

wn2 [HZ ]

0

wn5[HZ]

wn3 [HZ ]

10

x10-3

10

Estimated Actual value

∧

wn6 [HZ ]

Fig. 31.3 Impulse response of the six DOF system (sixth floor acceleration measurement) and its FRF (nominal system)

1 0

0

2

t[s]

4 t[s]

6

0 0

2

4 t[s]

6

0

t[s]

Fig. 31.4 Online estimation of system modal parameters

The new system modal parameters are shown in Table 31.2, where we can notice the changes on the natural frequencies and damping ratios. Notice that this induced fail does not imply obvious changes on the system dynamic behavior, however, the proposed identifier can detect them in a fast and online fashion, as shown in Fig. 31.6. The second case study consist on adding a 0.496 kg mass on the third floor only one second after the online estimation of the modal parameters from the nominal structure. This is shown in Fig. 31.7. Where we show the mass added to the third floor and the system time response after suddenly applying the mass and, therefore causing a new impact response of the structure.

256

L.G. Trujillo-Franco et al.

Fig. 31.5 Loose screw detail and impulse response of damaged structure Table 31.2 Change of the system modal parameters with a loose screw Frequency (Hz) Nominal 1.148 3.39 5.44 7.16 8.53 9.34

x10

2

4

6

0

1

2

3

4

5

0

2

4

6

0

2

4

1

2

3

4

5

6

5

1 0

0 0

2

4 t[s]

6

2

4

z3

2 0

0

1

2

3

4

5

6

t[s]

0

2

4 t[s]

6

2 2

4

6

2

4

6

4

6

2 0 0 -3 3 x 10

6

4

z4

∧

wn1

10

0 -3 x10

-3

0 0 -3 x 10 4

6

5 0

0

2 ∧

0 -3 x10

z1

∧

0

wn4

Normal Loose screw

∧

wn2

z5

5

10

1 0

6

%Change 13.33 5.182 11.11 10.52 5.55 35.294

x 10 4

z2

5 0

0

-3

2

5 0

wn5

10 ∧

10

Damping ratio (%) Nominal Loose screw 0.15 0.13 0.39 0.37 0.18 0.20 0.19 0.21 0.18 0.17 0.17 0.11

%Change 5.92 0.29 7.53 1.53 3.51 3.96

z6

Nominal loose screw

∧

wn6

15

Loose screw 1.08 3.38 5.03 7.05 8.23 8.97

wn3

Mode 1 2 3 4 5 6

2 1 0

0

2 t[s]

Fig. 31.6 Online modal parameters identification of structure with a loose screw

The fast and effective online modal parameters estimation is shown in Fig. 31.8 while the comparison of the new structure modal parameters is reported in Table 31.3. We intentionally leave idle the plotting part of the monitoring system for a second in order to plot only coherent parameters, that is, we wait for the algebraic identifier to finish the required estimations.

31.5 Conclusions It is proposed the use of an algebraic identification approach for the on-line monitoring and estimation of modal parameters for multiple degrees-of-freedom mechanical systems. In this application, we have considered that only measurements of one floor acceleration is available for the identification scheme implementation. The algebraic modal parameter identification

31 Online Systems Parameters Identification for Structural Monitoring Using Algebraic Techniques

257

Fig. 31.7 Mass change induced on the third floor after 3 s

5

6

z6 2

0

4

1

2

3

4

5

10

∧

0

1

2

3

4

5

6

2

4

0.02 0

0

1

2

3

4

5

6

0 0 -3 x10 5

2 1 0

2

4

0

1

2

t[s]

3

4

5

6

t[s]

0 0

0.01 0 0 -3 x10 4

6

5

∧

5 0

6

0

2

4

6

2

4

6

4

6

2 0 0

6

0.04

z4

0

wn1 [Hz]

0

0

6

∧

10

0

z2

4

z1

3

Mass change Normal

1

z5

2

z3

2 5

∧

1

wn2 [Hz]

0

∧

wn5 [Hz]

0

wn4 [Hz]

x10

10 wn3 [Hz]

10

∧

wn6 [Hz]

-3

Normal Mass change

2

t[s]

4

6

0.02 0

0

2

t[s]

Fig. 31.8 Online modal parameters identification when the structure has a mass change on the third floor

Table 31.3 Change of the system modal parameters with a mass added to the third floor Mode 1 2 3 4 5 6

Frequency (Hz) Nominal 1.148 3.39 5.44 7.16 8.53 9.34

With mass added 0.97 3.35 4.45 6.72 7.78 8.35

%Change 15.5 1.17 18.19 6.145 8.79 10.59

Damping ratio (%) Nominal With mass added 0.15 0.87 0.39 1.5 0.18 0.2 0.19 0.2 0.18 1.9 0.17 0.2

%Change 480 284 11.11 5.26 955 17.64

scheme was tested for a 6-DOF mechanical system excited by an impact excitation for illustrative purposes. In general, the simulation and experimental results show a satisfactory performance of the proposed identification approach with fast and effective estimations very promising for online applications as future operational modal analysis schemes and adaptive vibration absorption.

258

L.G. Trujillo-Franco et al.

References 1. Yang, Y., Nagarajaiah, S.: Output-only modal identification with limited sensors using sparse component analysis. J. Sound Vib. 332, 4741–4765 (2013) 2. Brincker, R., Ventura, C.E.: Introduction to Operational Modal Analysis. Wiley, Chichester (2015) 3. Golub, G., Van Loan, C.: An analysis of the total least squares problem. SIAM J. Numer. Anal. 17, 883–893 (1980) 4. Beltran-Carbajal, F., Silva-Navarro, G., Trujillo-Franco, L.G.: Evaluation of on-line algebraic modal parameter identification methods. In: Proceedings of the 32nd International Modal Analysis Conference (IMAC XXXII), vol. 8, pp. 145–152 (2014) 5. Beltran-Carbajal, F., Silva-Navarro, G., Trujillo-Franco, L.G.: Adaptive-like vibration control in a three-story building-like structure with a PZT stack actuator. In: Proceedings of the 33rd International Modal Analysis Conference (IMAC XXXIII), vol. 10, pp. 123–131 (2015) 6. Beltran-Carbajal, F., Silva-Navarro, G., Chávez-Conde, E.: Design of active vibration absorbers using on line estimation of parameters and signals. In: Beltrán-Carbajal, F. (ed.) Vibration Analysis and Control. New Trends and Developments. InTech, Croatia (2011) 7. Beltran-Carbajal, F., Silva-Navarro, G.: Algebraic parameter identification of multi-degree-of-freedom vibrating mechanical systems. In: Proceedings of the 20th International Congress on Sound and Vibration (ICSV20), Bangkok, Thailand, pp. 1–8 (2013) 8. Soderstrom, T., Stoica, P.: System Identification. Prentice-Hall, New York (1989) 9. Beltran-Carbajal, F., Silva-Navarro, G.: Adaptive-like vibration control in mechanical systems with unknown parameters and signals. Asian J. Control. 15, 1613–1626 (2013). doi:10.1002/asjc.7272013

Chapter 32

Structural Vibration Control Using High Strength and Damping Capacity Shape Memory Alloys Soheil Saedi, Farzad S. Dizaji, Osman E. Ozbulut, and Haluk E. Karaca

Abstract Designing structures to withstand dynamic environmental hazards such as earthquakes, strong winds, and hurricanes is of primary concern for civil engineers. In addition, recent advances in architectural forms, structural systems, and high performance materials have enabled the design of very slender and lightweight structures. These flexible structures are susceptible to be exposed to high levels of vibrations under strong winds and earthquakes, which may lead to structural damage and potential failure. Over the past two decades, shape memory alloys (SMAs) have emerged as a smart material that can be used in passive vibration control devices for energy dissipating and re-centering purposes. However, the quantity of equivalent viscous damping provided by superelastic NiTi SMA wires or bars is not sufficient to render the use of SMAs as the sole damping device implemented in a tall structure subjected to severe dynamic loadings. This study explores the performance of recently developed NiTiHfPd alloys that have very high strength, high dissipation/damping capacity, good cyclic stability and large operating temperature for vibration control applications. In particular, superelastic response of single crystal and polycrystalline NiTiHfPd alloys were investigated systematically to reveal the effects of composition alteration and heat treatments. The design of a novel passive vibration control device by utilizing the superelastic effect of NiTiHfPd SMAs under compression was illustrated. Keywords Shape memory alloys • Passive control • Damper • Vibrations • Earthquake

32.1 Introduction To mitigate the adverse effects of dynamic environmental hazards such as earthquakes, strong winds, and hurricanes and achieve a more resilient design under dynamic loads, various passive, active and semi-active control devices have been proposed and developed [1]. The most widely tested and commonly implemented strategies have been passive devices as they are reliable, require no external power, and never destabilize the structure. However, designing buildings with passive systems to provide performance improvement over all ranges of earthquake and wind excitation is a challenging task. A number of passive energy dissipation systems have been developed to mitigate damaging effects of natural hazards on structures [2]. Passive energy dissipation devices can be grouped into two main categories: hysteretic devices and ratedependent devices. Examples of hysteretic devices include metallic yielding devices and friction devices. Energy dissipation in hysteretic devices depends primarily on relative displacements within the device. These devices add initial stiffness until yielding or slip occurs and provide hysteretic energy dissipation. However, they do not provide sufficient damping at small vibrations caused by wind excitation or frequent seismic events while increase the forces and accelerations on the structure due to high stiffness. Furthermore, metallic devices usually have a limited number of working cycles and may require replacement after a strong event and friction devices may lead to permanent deformations if no restoring force mechanism is provided. Examples of rate-dependent devices include fluid viscous dampers and viscoelastic dampers. The energy dissipation capacity of these devices depends on the velocity across the device [2]. The viscous damper can provide high force and damping capabilities while the viscoelastic dampers have low force and displacement capacity. In general, rate-dependent devices can dissipate energy at all magnitudes of motion but do not possess the same capacity for energy dissipation as hysteretic devices [3]. Note that these devices also do not have self-centering property to return the structure to its original position.

S. Saedi • H.E. Karaca Department of Mechanical Engineering, University of Kentucky, Lexington, KY, USA F.S. Dizaji • O.E. Ozbulut () Department of Civil and Environmental Engineering, University of Virginia, Charlottesville, VA, USA e-mail: [email protected] © The Society for Experimental Mechanics, Inc. 2017 J. Caicedo, S. Pakzad (eds.), Dynamics of Civil Structures, Volume 2, Conference Proceedings of the Society for Experimental Mechanics Series, DOI 10.1007/978-3-319-54777-0_32

259

260

S. Saedi et al. 1600

NiTiHfPd (under compression)

1400

Stress (MPa)

1200

NiTi (under tension)

1000 800 600 400 200 0

0

1

2

3 4 Strain (%)

5

6

7

Fig. 32.1 Strain–stress diagrams for NiTi and NiTiHfPd

Numerous efforts have been carried out to develop new structural systems that can provide stable energy dissipation with full self-centering capabilities [4]. Due to its inherent and excellent re-centering ability, high corrosion and fatigue resistance, several researchers have proposed shape memory alloy (SMA)-based seismic protection systems including, but not limited to, SMA-based bracing systems [5–7]; SMA-based dampers [8–11]; and SMA-based isolation systems [12–16]. Among various SMA compositions, the NiTi alloy has been the most widely studied material for SMA-based dampers. Since NiTi SMAs does not have sufficient energy dissipation capacity to render the use of SMAs as the sole damping device, damping in those devices was supplemented through pre-tensioning SMA elements [17], incorporating a frictional device [18–20], energy-absorbing steel struts [21], viscoelastic device [22, 23], or buckling-restrained braces [24]. A recently developed class of SMAs that consists of NiTiHfPd provides large energy dissipation and high stresses [25] and can overcome the deficiency of currently available SMAs. Figure 32.1 presents typical strain-stress diagrams of NiTi and NiTiHfPd SMAs at room temperature. It can be seen that the area captured within the hysteresis loop, which signifies the dissipated energy, for NiTiHfPd is considerably larger than that of NiTi. The strength of NiTiHfPd is also significantly higher than that of NiTi. Thus, the required cross-sectional area of the SMA elements to develop the design force of a damping device will be considerably smaller for NiTiHfPd. A vibration control device based on NiTiHfPd will avoid employing additional energy dissipation unit and/or requiring large quantities of material due to high energy dissipation capabilities and strength of NiTiHfPd. This study discusses experimental characterization of NiTiHfPd SMAs and conceptual design of a simple but efficient NiTiHfPd SMA-based energy dissipating and re-centering device, which avoids extra fabrication and material costs.

32.2 High Damping and High Strength NiTiHfPd SMAs NiTi SMAs have been studied and employed in a variety of applications due to their superior shape memory properties, corrosion resistance, biocompatibility, ductility, etc. However, they have certain limitations that restrict their functionality for some applications. Alloying is one of the most effective ways to engineer properties and overcome these problems. For instance, an SMA material with good damping capacity needs to have large mechanical hysteresis and transformation strain at high-stress levels to be employed as dampers. While the mechanical hysteresis and damping capacity of binary NiTi alloys has been reported to be 200–400 MPa and 16 J cm3 respectively [26], with the addition of Nb it can be increased to 500– 600 MPa and 38 J cm3 [27]. The addition of elements like Zr, Hf, and Pd also has been studied to improve strength, wear resistance and damping capacity of SMAs. Due to its low cost, medium ductility and higher work output, the addition of Hf to NiTi binary found to be the promising [28, 29]. However, one of the main challenges that should be addressed before

32 Structural Vibration Control Using High Strength and Damping Capacity Shape Memory Alloys

261

NiTiHf alloys can be used in practical applications is their brittleness. Therefore, the addition of a variety of quaternary elements to NiTiHf has been explored in the literature. For example, the addition of Cu to NiTiHf has been found to shrink thermal hysteresis, improve thermal stability, ductility and two-way shape memory effect but no other substantial change has been observed for other shape memory properties of the alloy [30, 31]. The addition of Pd as a quaternary element to NiTiHf alloy has resulted in a significant improvement in the ductility, adjustment of TTs or enhance the shape memory behavior. Toward this goal, Pd can be added in the expense of Ti to increase the transformation temperatures or at the expense of Ni improves superelasticity. Single crystalline NiTiHfPd alloy has been reported to show huge mechanical hysteresis of up to 1270 MPa, superelastic response under extremely high-stress levels (2.5 GPa), and damping capacity of 44 J cm3 [25, 32]. Even polycrystalline NiTiHfPd alloys can generate high work outputs of 32–35 J cm3 (up to 120 ı C), which are considerably higher than NiTi and Ni. The ultra high-strength, and the exceptional damping capacity and practical ability to demonstrate superelastic behavior at room temperature makes NiTiHfPd SMAs a great candidate for applications that require high damping. Beside alloying, other most common methods to improve the shape memory and mechanical properties of SMAs are thermomechanical processing (e.g. cold working and post-annealing), precipitate formation, and grain refinement of polycrystalline alloys. However, thermal treatments seem to the most practical and cost-effective method amongst the others. It has been shown that by the formation of precipitates through aging, the strength of the matrix and required critical stress are increased which results in better shape memory response, the fatigue life, and cyclic stability. It should be noted that the strengthening ability of the precipitates depends on the size, volume fraction, antiparticle spacing, and coherency of these particles [33].

32.3 Material Characterization For experiment characterization, Ni-rich single [111] oriented single crystal Ni45.3 Ti29.7 Hf20 Pd5 (at %) and polycrystal Ni45.7 Ti29.3 Hf20 Pd5 (at %) alloys obtained from NASA. Perkin-Elmer Pyris 1 differential scanning calorimetry (DSC), was used to determine the transformation temperatures. Heating–cooling rate was adjusted to 10 ı C/min. Aging of the samples was performed using Lindberg/Blue M BF514541 furnace in air atmosphere and then water quenched. The mechanical experiments were performed in an MTS Landmark servo-hydraulic test platform (100 kN) on 8  4  4 mm3 compression specimens cut by electro-discharge machining. The strain was measured by an MTS high-temperature extensometer with a gage length of 12 mm attached to compression grip faces. A strain rate of 104 s1 was employed during loading of the specimens, while unloading was performed at a rate of 100 N/s. Heating of the specimens was occurred by means of mica band heaters retrofitted to the compression grips, and cooling was achieved through internal liquid nitrogen flow in the compression grips. A heating–cooling rate of 10 ı C/min was applied during testing, using an Omega CN8200 series PID temperature controller. K-type thermocouples attached to the test specimens and the compression grips provided real-time temperature feedback.

32.4 Results and Discussion Figure 32.2 demonstrates the superelastic response of [111] oriented single crystal NiTiHfPd. For superelastic experiments, samples were heated up to a temperature well above Af and loaded up to 2, 4 and 6% strain limit successively and then unloaded. The tests were performed at room temperature since the sample was well-above Af temperature at this temperature. Next sample was loaded until failure to examine the ductility. The critical stress for stress induced martensitic transformation was occurred at about 1300 MPa, and therefore 2% loading did not reach to this point and sample showed a linear behavior. After 4 and 6% deformation, the sample was capable to show full strain recovery at room temperature and mechanical hysteresis was more than 700 MPa at 6% deformation. However, when sample loaded further, it was failed at 1955 MPa and 7.2% deformation without showing the second plateau (for plastic deformation) indicating high brittleness of the sample. As mentioned earlier, it is possible to tailor mechanical properties significantly using thermal treatments. The same alloy has been showed even higher strength as well as higher mechanical hysteresis in homogenized condition (1050 ı C–4 h) shown at Fig. 32.3. The homogenized sample shows ultra-high strength with a very wide superelastic window covering from 30 to 50 ı C. The critical stress for transformation was more than 1500 MPa at 10 ı C. The stress hysteresis was changing from 1270 to 900 MPa and decreasing as a function of critical stress of temperature [25], which is the highest reported hysteresis in NiTi based SMAs literature.

262

S. Saedi et al.

Fig. 32.2 Superelastic response of [111] oriented single crystal NiTiHfPd—as-received

Fig. 32.3 Superelastic response of [111] oriented single crystal NiTiHfPd in homogenized condition (extracted from [25])

Since single crystal alloys are costlier and difficult to fabricate, polycrystal samples are much more demanding despite their lower strength. However, using precipitation formation through post processing heat treatments, it is possible to improve their characteristics [34]. Figure 32.4 displays the effects of aging time and temperatures on TTs of polycrystal NiTiHfPd samples, aged from 0.5 to 3 h at 350–550 ı C. Three heating-cooling cycles have been performed on each sample to find the stabilized TTs. The As-received condition has been included in every figure to ease the comparison. From the Fig. 32.4a, it can be seen that TTs have decreased as a function of time when they aged at 350 ı C. A similar trend was observed for samples aged at 450 and 550 ı C in Fig. 32.4b, c, however, the decrease is minor. Another important observation is aging at 450 and 550 ı C even for 0.5 h causes a huge shift in TTs when compared to the as-received condition. Figure 32.4d compares the samples aged for 3 h at different temperatures. On contrary, TTs were increased as a function of temperature. While Ms and Af temperatures of 350 ı C–3 h aged samples were 16 and 27 ı C respectively after increasing the aging temperature to 550 ı C, they were found to be 189 and 216 ı C. The increases in TTs of Ni-rich alloys can be attributed to the formation of large Ni-rich precipitates. These precipitates deplete the matrix from Ni, therefore, TTs are increased [35]. As the aging temperature increases, they grow larger, their volume fraction and inter-particle distance increases, which result in more Ni depletion in the matrix, consequently TTs maintain the increasing trend. Furthermore, internal stress and

32 Structural Vibration Control Using High Strength and Damping Capacity Shape Memory Alloys

263

Fig. 32.4 DSC graphs of poly crustal NiTiHfPd aged at (a) 350 ı C, (b) 450 ı C, (c) 550 ı C and (d) 3 h aging comparison

coherency of the formed precipitates are important factors affecting the TTs. The mismatch between matrix and precipitate lattice parameters may cause internal stresses around precipitates that can alter the TTs. Therefore, increasing of the formed precipitates coherency with aging temperatures that causes higher internal stresses can be named as the second reason for increase in TTs. The drop of TTs for 350 ı C aging again can be attributed to the formation of precipitates with very small size and short interparticle distance. Thus, the resistance for martensite nucleation was increased and formation of martensite required additional energy change, which in turn required further undercooling in the course of the forward transformation. Figure 32.5a, b show the superelastic response of NiTiHfPd alloy in as-received and 450 ı C–10 h aged conditions, respectively. To avoid further plastic deformation aged sample was loaded up to 5% at last cycle. It can be seen from Fig. 32.4a that in as-received condition sample can display only 2% full recovery and further deformation results in irrecoverable strain. On the other hand, 450 ı C–10 h aged sample as shown in Fig. 32.4b showed almost full strain recovery even after 4% deformation. The irrecoverable strain was only 0.65 after 5% deformation. When these condition are compared, it is observed that even though critical stress for stress-induced martensitic transformation occurred at higher stress level for asreceived samples than aged samples, strain hardening with a sharp slope was observed after plateau, which causes a plastic deformation of the sample. It is noteworthy that as-received sample also shows a very low energy dissipation, resulting in very narrow stress hysteresis while aged sample showed about 300 MPa stress hysteresis after 2% deformation, which was increased to about 600 MPa when deformed 4%.

264

S. Saedi et al.

Fig. 32.5 Superelastic response of NiTiHfPd (a) As-received, (b) aged 450 ı C–10 h

Fig. 32.6 Plan view of SMARD at its original and displaced positions

32.5 Shape Memory Alloy-Based Re-centering Damper By exploiting the advantageous characteristics of NiTiHfPd SMAs, a new passive vibration device is proposed in this study. The device is named as Shape Memory Alloy-based Re-centering Damper or SMARD. Figure 32.6 shows plan and longitudinal view of the SMARD at its original and displaced positions to describe its operating principle. A threedimensional rendering of the device is also shown in the Fig. 32.7. The SMARD comprises a SMA-Spring assembly (rendered in yellow), a piston (gray), and an enclosure (black) with rigidly attached baffle plates. The high force steel springs between SMA groups ensure large displacements under load while simultaneously transmitting the force between SMA groups. Voids in the centers of the baffle plates allow the piston to pass through freely while arresting the otherwise free-traveling SMA-Spring assembly. Note that NiTiHfPd SMAs exhibit better superelastic response under compression. Therefore, the device is designed such that whether the piston is drawn out of or forced into the device, the SMA-Spring assembly will undergo compression. Due to the inherent superelastic behavior of NiTiHfPd SMAs, the SMARD can exhibit excellent re-centering ability with enhanced energy dissipation capacity. In addition, SMA bars with up to four different diameters can be incorporated to the device. That will enable device to produce a damping force that ensures sufficient energy dissipation for various levels of excitation. The SMARD is illustrated with two different diameter bars (SMA1 and SMA2) in Fig. 32.6. As a result of arrangement of SMA bars in the device and inherent re-centering behavior of SMAs, a properly designed SMARD will exhibit a re-centering and energy dissipating behavior during all modes of vibrations under both wind and earthquake excitations. During a weak to moderate event, the SMA bars with lower stiffness will experience relatively larger compressive strains and dissipate the energy while the stiffer SMA bars will undergo small strains. During a moderate to strong event, both SMA groups will engage and dissipate large amounts of energy, and the device will provide a larger damping force. During an extreme event, the SMARD will stiffen due to second stiffening of SMA bars under large strains. This additional stiffening will avoid potentially catastrophic structural displacements even under unforeseen strong ground motions.

32 Structural Vibration Control Using High Strength and Damping Capacity Shape Memory Alloys

265

Fig. 32.7 Three-dimensional rendering of SMARD

32.6 Conclusions This study proposes a new passive vibration control device that incorporates high performance superelastic shape memory alloys. These new superelastic SMAs consist of recently developed NiTiHfPd alloys that have very high strength, high dissipation/damping capacity, good cyclic stability and large operating temperature. Superelastic response of a single crystal NiTiHfPd was briefly discussed. Then, a systematic study was conducted to evaluate the mechanical response of polycrsytal NiTiHfPd SMAs. In particular, differential scanning calorimetry (DSC) was used to determine the effect of aging time and temperature on transformation temperatures. Compressive tests were conducted to reveal the superelastic response of polycrsytal NiTiHfPd alloy in as-received and 450 ı C–10 h aged conditions. The results obtained from these initial experimental efforts suggest that polycrystal NiTiHfPd SMAs can also exhibit a superelastic response at very high stress levels and with a large hysteresis loop. However, further investigations will be conducted to determine the effects of heat treatments on transformation temperatures and provide the ability to tailor the microstructure to get superelastic behavior at a temperature range of 20 to 50 ı C. Once the material characteristics of polycrystalline NiTiHfPd are optimized, a prototype of the proposed SMARD device will be fabricated and tested. Acknowledgements This material is based upon the work supported by the National Science Foundation under Grant Number CMMI-1538770.

References 1. Symans, M.D., Charney, F.A., Whittaker, A.S., Constantinou, M.C., Kircher, C.A., Johnson, M.W., McNamara, R.J.: Energy dissipation systems for seismic applications: current practice and recent developments. J. Struct. Eng. 134(1), 3–21 (2008) 2. Soong, T.T., Dargush, G.F.: Passive Energy Dissipation System Structural Engineering. Wiley, Chichester (1997) 3. Marshall, J.D., Charney, F.A.: Seismic response of steel frame structures with hybrid passive control systems. Earthq. Eng. Struct. Dyn. 41(4), 715–733 (2012) 4. Chancellor, N.B., Eatherton, M.R., Roke, D.A., Akba¸s, T.: Self-centering seismic lateral force resisting systems: high performance structures for the city of tomorrow. Buildings. 4(3), 520–548 (2014) 5. Ozbulut, O.E., Mir, C., Moroni, M.O., Sarrazin, M., Roschke, P.N.: A fuzzy model of superelastic shape memory alloys for vibration control in civil engineering applications. Smart Mater. Struct. 16, 818–829 (2007) 6. Qiu, C.X., Zhu, S.: Performance-based seismic design of self-centering steel frames with SMA-based braces. Eng. Struct. 130, 67–82 (2017) 7. Gao, N., Jeon, J.S., Hodgson, D.E., DesRoches, R.: An innovative seismic bracing system based on a superelastic shape memory alloy ring. Smart Mater. Struct. 25(5), 055030 (2016) 8. Ozbulut, O.E., Roschke, P.N., Lin, P.Y., Loh, C.H.: GA-based optimum design of a shape memory alloy device for seismic response mitigation. Smart Mater. Struct. 19, 065004 (2010) 9. Ozbulut, O.E., Hurlebaus, S.: Application of an SMA-based hybrid control device to 20-story nonlinear benchmark building. Earthq. Eng. Struct. Dyn. 41, 1831–1843 (2012) 10. Mishra, S.K., Gur, S., Chakraborty, S.: An improved tuned mass damper (SMA-TMD) assisted by a shape memory alloy spring. Smart Mater. Struct. 22(9), 095016 (2013)

266

S. Saedi et al.

11. Parulekar, Y.M., Kiran, A.R., Reddy, G.R., Singh, R.K., Vaze, K.K.: Shake table tests and analytical simulations of a steel structure with shape memory alloy dampers. Smart Mater. Struct. 23(12), 125002 (2014) 12. Ozbulut, O.E., Hurlebaus, S.: A comparative study on seismic performance of superelastic-friction base isolators against near-field earthquakes. Earthq. Spectra. 28, 1147–1163 (2012) 13. Ozbulut, O.E., Hurlebaus, S.: Energy-balance assessment of shape memory alloy-based seismic isolation devices. Smart Struct. Syst. 8, 399–412 (2011) 14. Ozbulut, O.E., Hurlebaus, S.: Seismic assessment of bridge structures isolated by a shape memory alloy/rubber-based isolation system. Smart Mater. Struct. 20, 015003 (2011) 15. Dezfuli, F.H., Alam, M.S.: Shape memory alloy wire-based smart natural rubber bearing. Smart Mater. Struct. 22(4), 045013 (2013) 16. Ghodke, S., Jangid, R.S.: Influence of high austenite stiffness of shape memory alloy on the response of base isolated benchmark building. Struct. Control Health Monit. 24, e1867 (2016). doi:10.1002/stc.1867 17. Dolce, M., Cardone, D., Marnetto, R.: Implementation and testing of passive control devices based on shape memory alloys. Earthq. Eng. Struct. Dyn. 29, 945–968 (2000) 18. Zhu, S., Zhang, Y.: Seismic behaviour of self-centring braced frame buildings with reusable hysteretic damping brace. Earthq. Eng. Struct. Dyn. 36, 1329–1346 (2007) 19. Ozbulut, O.E., Hurlebaus, S.: Re-centering variable friction device for vibration control of structures subjected to near-field earthquakes. Mech. Syst. Signal Process. 25, 2849–2862 (2011) 20. Ozbulut, O.E., Bitaraf, M., Hurlebaus, S.: Adaptive control of base-isolated structures against near-field earthquakes using variable friction dampers. Eng. Struct. 33, 3143–3154 (2011) 21. Yang, C.W., DesRoches, R., Leon, R.T.: Design and analysis of braced frames with shape memory alloy and energy-absorbing hybrid devices. Eng. Struct. 32, 498–507 (2010) 22. Silwal, B., Michael, R.J., Ozbulut, O.E.: A superelastic viscous damper for enhanced seismic performance of steel frame structures. Eng. Struct. 105, 152–164 (2015) 23. Silwal, B., Ozbulut, O.E., Michael, R.J.: Seismic collapse evaluation of steel moment resisting frames with superelastic viscous damper. J. Constr. Steel Res. 126, 26–36 (2016) 24. Miller, D.J., Fahnestock, L.A., Eatherton, M.R.: Development and experimental validation of a nickel–titanium shape memory alloy selfcentering buckling-restrained brace. Eng. Struct. 40, 288–298 (2012) 25. Karaca, H., et al.: Superelastic response and damping capacity of ultrahigh-strength [111]-oriented NiTiHfPd single crystals. Scripta Mater. 67(5), 447–450 (2012) 26. Otsuka, K., Ren, X.: Physical metallurgy of Ti–Ni-based shape memory alloys. Prog. Mater. Sci. 50(5), 511–678 (2005) 27. Tanaka, Y., et al.: Ferrous polycrystalline shape-memory alloy showing huge superelasticity. Science. 327(5972), 1488–1490 (2010) 28. Karaca, H., et al.: Compressive response of nickel-rich NiTiHf high-temperature shape memory single crystals along the [111] orientation. Scripta Mater. 65(7), 577–580 (2011) 29. Karaca, H., et al.: NiTiHf-based shape memory alloys. Mater. Sci. Technol. 30(13), 1530–1544 (2014) 30. Liang, X., et al.: Thermal cycling stability and two-way shape memory effect of Ni–Cu–Ti–Hf alloys. Solid State Commun. 119(6), 381–385 (2001) 31. Meng, X.L., et al.: Martensite structure in Ti–Ni–Hf–Cu quaternary alloy ribbons containing (Ti,Hf)2Ni precipitates. Acta Mater. 58(10), 3751–3763 (2010) 32. Karaca, H., et al.: Effects of aging on [111] oriented NiTiHfPd single crystals under compression. Scripta Mater. 67(7), 728–731 (2012) 33. Saedi, S., et al.: The influence of heat treatment on the thermomechanical response of Ni-rich NiTi alloys manufactured by selective laser melting. J. Alloys Compd. 677, 204–210 (2016) 34. Turabi, A.S., et al.: Experimental characterization of shape memory alloys. In: Elahinia, M. (ed.) Shape Memory Alloy Actuators: Design, Fabrication and Experimental Evaluation, pp. 239–277. Wiley (2015) 35. Saedi, S., et al.: Thermomechanical characterization of Ni-rich NiTi fabricated by selective laser melting. Smart Mater. Struct. 25(3), 035005 (2016)

Chapter 33

Comparative Study on Modal Identification of a 10 Story RC Structure Using Free, Ambient and Forced Vibration Data Seyedsina Yousefianmoghadam, Andreas Stavridis, and Babak Moaveni

Abstract In this study the modal parameters of a ten-story reinforced concrete (RC) building are estimated from recordings of free, ambient, and forced vibrations, using the eigensystem realization algorithm (ERA), the natural excitation technique combined with the eigensystem realization algorithm (NExT-ERA), and peak-picking method, respectively. The structure vibrated freely after it was hit the by a jack hammer used to demolish selected perimeter walls in the second story. The forced vibrations were induced using a portable eccentric mass shaker installed on the second story of the structure. The results obtained from the three methods indicate that the free vibration recordings can provide an accurate estimation of the modal parameters of the structure at lower identification order and with reduced computational cost compared to the NExT-ERA method applied on the ambient vibration recordings. Keywords System identification • Impulse response • Dynamic testing • Comparative study • RC structure

33.1 Introduction Methods to assess the condition of existing structures using vibration-based diagnostic methods are proposed in a number of studies [1–3]. In these studies, the considered vibrations can be caused by ambient excitations such as wind, nearby vehicular or pedestrian traffic [4–11] or by controlled excitations induced by an impact hammer, a shaker, a shake table, a truck, or a jack [12–17]. In terms of the considered structures, these include actual buildings and bridges, as well as largescale laboratory specimens, with the majority being laboratory specimens since the dynamic testing of actual structures is not common due to the practical implications and risks involved with such tests. This paper discusses the modal identification of free, ambient, and forced vibration tests on a 10-story reinforced concrete (RC) building at four damage states. The damage states were induced through the demolition of infill walls in the perimeter which was permitted since the structure was to be demolished. The demolition of the walls with an impact hammer induced high amplitude impulses followed by the free vibration of the building. Moreover, harmonic excitations were induced to the structure using a portable shaker. The dynamic properties of the building at each damage state are identified from the free vibration data using the eigensystem realization algorithm (ERA), the ambient vibrations using the natural excitation technique combined with the eigensystem realization algorithm (NExT-ERA), and the forced vibrations using the peakpicking method. The comparison of the dynamic properties obtained from the three types of excitation indicates that all of the methods can provide accurate results, with varying computational costs. Finally, a parametric study using the NExT-ERA method is also performed to investigate the effect of the number and location of selected reference channels on the estimated modal parameters.

S. Yousefianmoghadam () University at Buffalo, 116 Ketter Hall, Buffalo, NY 14260, USA e-mail: [email protected] A. Stavridis University at Buffalo, 224 Ketter Hall, Buffalo, NY 14260, USA B. Moaveni Tufts University, Medford, MA, USA © The Society for Experimental Mechanics, Inc. 2017 J. Caicedo, S. Pakzad (eds.), Dynamics of Civil Structures, Volume 2, Conference Proceedings of the Society for Experimental Mechanics Series, DOI 10.1007/978-3-319-54777-0_33

267

268

S. Yousefianmoghadam et al.

Fig. 33.1 Test structure (view from west side)

Fig. 33.2 Second floor plan view showing extension building and acceleration positions

33.2 Test Structure The structure, shown in Fig. 33.1, was a 10-story reinforced concrete (RC) structure with a slab-column structural system and RC infills in the exterior frames. The building was built in 1910s in Utica, NY and characterized by overall length of 48.8 m (160 ft), width of 24.4 m (80 ft), and height of 25.9 m (85 ft). The plan view of the second floor can be seen in Fig. 33.2. This is similar to the plan view of other floors. On the south side of the building, a five-story clay masonry structure was attached to it to house an elevator shaft and service rooms. More detailed information about the structure can be found in [18].

33.3 Dynamic Testing 33.3.1 Damage States During the testing sequence, six exterior RC walls were removed from the second and third stories of the structure to simulate damage. The walls were removed, two at a time, in three stages, as shown in Fig. 33.3a, introducing four damage states. The first damage state (DS0) was the condition of the structure at the beginning of the tests in which the exterior walls in the first, second, and fifth stories, in the north, west, and south sides were removed for demolition purposes. The second damage state

33 Comparative Study on Modal Identification of a 10 Story RC Structure Using Free, Ambient and Forced Vibration Data

269

Fig. 33.3 Wall removal and damage states. (a) Structure initial condition and damage states. (b) Wall removal using jack hammer

Fig. 33.4 Example of the time history responses for different excitation types recorded at 10th floor. (a) Ambient vibration. (b) Forced vibration (sine sweep). (c) Wall demolition in 2nd story

(DS1) resulted from the removal of two walls in the western exterior frame on the third story. The third state of damage (DS2) occurred from the removal of two walls in the northern exterior frame on the third story. The fourth and final damage state (DS3) occurred after the demolition of walls in the north-west corner of the second story. The wall removal was achieved using a jack hammer installed on an excavator as shown in Fig. 33.3b.

33.3.2 Testing Method and Sequence A series of dynamic tests were performed on the building using a mobile shaker bolted to the second floor slab in a location eccentric along the Y axis as shown in Fig. 33.2. The harmonic excitations induced by the shaker were sine sweeps, sine steps, and sine dwells around the identified natural frequencies. More details about the dynamic tests are available in [19]. This study focuses on the sine sweeps and the excitations due to the jack hammer impacts to the exterior RC walls of the structure during the demolition process. Figure 33.4 shows the sample response of the structure recorded at the 10th floor. It can be seen that accelerations induced during the demolition were significantly larger compared to those induced during the forced vibrations using the shaker.

270

S. Yousefianmoghadam et al.

33.3.3 Instrumentation The accelerations along the vertical and the two horizontal directions at two locations of every slab were recorded with force-balance accelerometers. As indicated in Figs. 33.2 and 33.3, the accelerometers were installed near the NW and SE corners on all stories except for the roof due to safety reasons. This configuration was selected so that the translational, as well as the torsional motion of the structure could be monitored. The data logger used in this study could record accelerations between 0.5 g and 2 g. The sampling rate was 200 Hz and all accelerometers were synchronized by GPS timing having less than 1 milli-second accuracy [19]. During the testing period, the data acquisition system was continuously recording for 56 h. Hence, the ambient vibrations of the building were recorded, as well as the forced vibrations and the free vibrations.

33.4 System Identification The Eigensystem Realization Algorithm (ERA) and the Natural Excitation Technique combined with the Eigensystem Realization Algorithm (NExT-ERA) are used to estimate the modal properties using the free vibration and ambient measurements respectively. The peak-picking method is used to obtain the modal properties of the building from the transfer function of the force vibrations due to the sine sweep excitations. The modal properties of the test structure are estimated at each damage state using all data sets. The free vibrations following each impulse were extracted from the recordings. They include, on average, 9 s of the response after the impulse. Since these excitations were induced during the wall demolition, only the free vibrations at the beginning and end of each wall demolition stage are considered here. At the beginning of each wall demolition phase, the vibrations following each hit of the hummer are attributed to the previous damage state while the vibrations towards the end of the demolition are attributed to the following damage state. Figure 33.5 shows an example of free vibration acceleration time history. A total of 60 free vibration sets are obtained. The obtained records are filtered in the frequency range between 0.2 and 20 Hz using a finite response filter of order 8192 and they are then down-sampled from 200 to 50 Hz. A block Hankel matrix of (40  200)  (50  signal duration) is used for the Eigensystem Realization Algorithm. Table 33.1 summarizes the cases considered here. In the case of ambient recordings, continuous acceleration recordings of 12 to 15 min are considered. The recordings are filtered and down sampled as those obtained from the free vibrations. The auto- and cross-correlation of signals are estimated as the inverse Fourier transformation of the power and cross spectral densities which are calculated using the Welch method [20]. The spectral densities are averaged over 72-s long windows with 50% overlap to mitigate the effects of noise while a Hanning function [21] is used to eliminate the effects of window leakage. To investigate the effect of the number and location of reference channels, six combinations of reference channels are considered in this study. Once the auto/cross correlations

Fig. 33.5 Acceleration time history of a free vibration measured on the 10th floor at the NE corner

33 Comparative Study on Modal Identification of a 10 Story RC Structure Using Free, Ambient and Forced Vibration Data

271

Table 33.1 Summary of free vibration measurements properties Damage state DS0 DS1 DS2 DS3

Number of recordings 12 12 17 19

Signal duration range (s) 6–25 8–14 6–10 5–12

Impulse direction Y Y X X

Max. acceleration range in all 40 channels (mg) 1.5–15 1.2–14.5 0.4–4.66 0.5–28

Table 33.2 Cases studied for NexT-ERA method # of reference channels 4 2 1 1 1 1

Maximum acceleration recorded among all channels (milli g)

Case # 1 2 3 4 5 6

18 16 14 12 10 8 6 4 2 0

Location of reference channels 10th floor: SW and NE corners 10th floor: SW corner 10th floor: SW corner 10th floor: SW corner 5th floor: SW corner 1st floor: SW corner

Reference channel measurement direction(s) X and Y X and Y X Y Y X

Mode 1 Mode 2 Mode 3 Mode 4

0

10

20

30

40

50

60

Identification order Fig. 33.6 Identification order dependency on maximum acceleration of the free vibration measurements in Y direction at DS0

are estimated, they are used to form a block Hankel matrix of (40  200)  ((50  72)  (# of reference channels)). Table 33.2 presents the cases considered in this study. For the forced vibration recordings the power spectral densities (PSD) of the acceleration measurements are estimated using the Welch method. The signals are averaged over Hamming windows of 8000 points and 50% window overlap. The PSDs are used to obtain the transfer functions. The acceleration measurements in X and Y directions at the SW corner of the 2nd floor are considered as the input excitation due to the proximity to the shaker.

33.4.1 Identification Orders and Stabilization Diagrams According to a previous study on system identification of the same structure [19], the first mode of the structure is identified at an order of 130 at DS0. Therefore, the system identification process for both ambient vibration and free vibration measurements is performed up to the order of 150. Figure 33.6 illustrates the identification order for the first four modes of the structure plotted against the maximum acceleration recorded among all of the 40 channels during the free vibration response at DS0 using the ERA method. Typically, identification orders as low as 10 are sufficient to identify the first two modes, and orders between 20 and 30 are sufficient for modes 3 and 4. It can be also observed in Fig. 33.6 that in the case of free vibrations no relation can be established between the identification order and the amplitude of the impulse excitations. The identification orders from the ambient vibration are compared to the average order needed in the analysis of the free vibration recordings at DS0 in Fig. 33.7. It can be observed that the identification orders are much higher using the ambient vibration data for modes 1–3 in all cases of reference channels considered for the first three modes, and in a few cases the modes are not identified at all. The difficulty in identifying the modes in lower system orders may be caused by processing of the signal used to convert the ambient vibration to free vibration as discussed in a previous section. This process is based on the assumption that the signal is broadband excitation which is not accurate. The unavoidable error in this simplifying

272

S. Yousefianmoghadam et al.

Demolition Average Case 1 Case 2 Not identified Case 3 Case 4 Case 5 Case 6 Not identified

140

Identification order

120 100 80 60 40 20 0

Mode 1

Mode 2

Mode 3

Mode 4

Fig. 33.7 Comparison of identification orders between the studied cases

assumption along with the computational error added to the system identification because of the conversion process can increase the identification order and required computational resources. Figure 33.7 also demonstrates the importance of the number and location of the reference channels in the modal identification of ambient vibration recordings using NExT-ERA method. When four, two, or one reference channels at the 10th floor are considered in cases 1, 2, and 4, respectively, the identification order slightly increases as the number of reference channels decreases. However, the computational time needed to run the identification algorithm increases significantly by increasing the number of reference channels as the Henkel matrix size is proportional to the number of channels. Case 5 which considers one reference channel at the fifth floor requires a higher identification order than case 4 which includes one reference channel in the 10th floor. This can be attributed to the fact that the higher floors experience more motion which increases the signal-to-noise ratio. In case 3, which considers one reference channel at the 10th floor measuring the acceleration in a direction perpendicular to the that of case 4 (X-direction), only mode 2 can be identified at DS0. This can also be observed in Table 33.3 which summarizes the identification order of the considered cases at all damage states. In all damage states, case 3 identifies modes in higher orders than those of case 4. As shown in Fig. 33.8, which illustrates the mode shapes of the structure, the reference channel selected for case 3 (X-direction in the South-West corner at the 10th floor) has a relatively small modal component in all modes except for mode 2. Therefore, if this channel is used as a reference, the relatively low signal-to-noise ratio in the free vibrations is produced from the NExT process. Case 6 which uses reference signal obtained from an accelerometer in the first floor at the direction with small modal components, combines the drawbacks of cases 3 and 5. This results in the highest identification orders among all the considered cases.

33.4.2 Modal Frequencies and Mode Shapes Table 33.3 also presents the identified natural frequencies of the building for all considered cases. The values match well between cases 1, 2, and 4 with maximum difference within 1%. In cases 3, 5, and 6 the results are more scattered; however, the difference is not considerable. One can note that the natural frequencies identified from the free vibration measurements are lower than those obtained from the recordings of ambient and forced vibrations. This difference can be attributed to the level of motion of the structure as reflected in the acceleration amplitude. As shown in Fig. 33.4, the acceleration of the structure after the jack hammer impulses has a higher amplitude than those resulted from forced and ambient vibrations. The higher amplitude causes opening of minor cracks which reduce the overall stiffness of the structure and therefore the identified natural frequencies. This can be also seen in Fig. 33.9 which presents the identified natural frequency at different damage states for the first two modes of the structure using the different data sets. In all cases, the identified frequencies indicate the same frequency change between the damage states. Figure 33.8 presents the mode shapes obtained from the average of the free vibration, ambient vibration in cases 1, 2, and 4, and the forced vibration measurements. The modal assurance criterion (MAC) [22] values are also calculated using

a

The mode was not identified

Identification Damage state Mode order DS0 1 44 2 36 3 110 4 36 DS1 1 14 2 42 3 42 4 34 DS2 1 22 2 16 3 8 4 40 DS3 1 48 2 24 3 6 4 6

1

IdentifiFrequency cation order (Hz) 2.26 44 3.34 36 4.69 108 7.37 36 2.24 14 3.33 42 4.71 44 7.37 38 2.19 22 3.25 16 4.71 8 7.40 46 2.11 48 3.10 24 4.65 6 7.30 6

2

Ambient vibration case

IdentifiFrequency cation order (Hz) 2.26 N/Ia 3.34 40 4.69 N/Ia 7.37 N/Ia 2.24 84 3.33 78 4.67 76 7.37 50 2.18 N/Ia 3.25 18 4.71 54 7.37 76 2.11 146 3.10 26 4.65 6 7.30 90

3 IdentifiFrequency cation order (Hz) N/Ia 46 3.31 36 N/Ia 106 N/Ia 36 2.26 14 3.30 42 4.73 42 7.31 48 N/Ia 22 3.25 14 4.68 6 7.39 42 2.10 40 3.10 26 4.65 6 7.25 10

4

Table 33.3 Summary NExT-ERA system identification on ambient vibration measurements

IdentifiFrequency cation order (Hz) 2.24 68 3.34 48 4.68 140 7.36 30 2.25 24 3.32 54 4.72 56 7.37 94 2.18 24 3.25 18 4.71 14 7.38 28 2.12 56 3.10 28 4.65 6 7.33 10

5 IdentifiFrequency cation order (Hz) 2.26 N/Ia 3.35 N/Ia 4.69 N/Ia 7.32 66 2.24 88 3.34 144 4.72 80 7.36 N/Ia 2.18 88 3.25 54 4.71 90 7.34 68 2.12 N/Ia 3.11 112 4.65 6 7.32 8

6

Average Frequency identification order (Hz) N/Ia 11 N/Ia 16 N/Ia 44 7.35 33 2.25 13 3.33 24 4.71 35 N/Ia 35 2.20 21 3.25 20 4.71 30 7.33 46 N/Ia 22 3.11 27 4.66 19 7.33 30

Free vibration Average frequency (Hz) 2.23 3.31 4.65 7.35 2.21 3.26 4.70 7.32 2.16 3.20 4.67 7.38 2.07 3.05 4.64 7.23

Average frequency (Hz) 2.23 3.32 4.66 N/A 2.23 3.30 4.71 N/A 2.17 3.24 4.68 N/A 2.09 3.08 4.66 N/A

Forced vibration

33 Comparative Study on Modal Identification of a 10 Story RC Structure Using Free, Ambient and Forced Vibration Data 273

274

S. Yousefianmoghadam et al.

Frequency (Hz)

Fig. 33.8 Mode shapes of the structure at DS0 3.4 3.2 3.0 2.8 2.6 2.4 2.2 2.0

Mode1-Ambient vibration Mode1-Demolition Mode1-Forced vibration Mode2-Ambient vibration Mode2-Demolition

DS0

DS1

DS2

DS3

Fig. 33.9 Identified natural frequencies at different damage states

the 40-component mode shape vectors between the average mode shapes estimated using different sources of excitation described above. It can be seen from the figure that the mode shapes are in excellent agreement with the MAC value of 0.98 or higher.

33.5 Conclusions and Remarks A comparative study between the system identification methods, ERA, NExT-ERA, and peak picking using the free vibration, ambient vibration, and forced vibration data respectively is presented here. The measurements were obtained from dynamic tests of a 10-story RC building which was damaged through the demolition of selected perimeter walls using a jack hammer. The identification orders used in the ERA method applied on the free vibration measurements are significantly lower than those of NExT-ERA method applied on the ambient vibration recordings. This difference may occur because of the additional processing of the ambient vibration signals to obtain the free vibration signal (NExT) which may add computational errors to the identification process.

33 Comparative Study on Modal Identification of a 10 Story RC Structure Using Free, Ambient and Forced Vibration Data

275

The study also investigates the effect of the number and location of the reference channels on the system identification results. Increasing the number of the reference channels results in lower identification orders but no significant difference is observed in the values of the identified natural frequencies. However, the increased number of reference channels requires significantly more computational time. The location of the reference channels affects drastically the results of the modal identification. Hence, when a reference channel measuring the acceleration at a location and direction which has a low participation in mode shapes of the structure is used, the system identification is not accurate as certain modes may not be identified. Moreover, channels with higher signal-to-noise ratio result in more accurate system identification. The identified frequencies from the three types of measurements indicate the same decreasing trend between the damage states confirming the ability of the all methods to identify the changes in the dynamic properties of the system. It is also observed that higher amplitude of the excitation results in lower identified frequencies as indicated by the comparison of the frequencies obtained from the free vibrations following the impulses induced by the jack hammer and the frequencies identified from the ambient vibration of the building. Finally, using the ERA method accurate results can be obtained at less computational time. Moreover, this method does not suffer from issues related to the choice of reference channels. Based on the observations, at this excitation level, the amplitude of impulse is independent of the modal identification and even low amplitude non-destructive impulses can induce free vibrations to be used for ERA method for healthy and in service structures. Acknowledgements The study presented here is part of a project supported by the National Science Foundation (Award No. 1430180). The collaboration of NEES@UCLA in conducting the experiments is sincerely acknowledged. The authors would also like to thank the New York State Department of transportation (NYSDOT) personnel and director Andrew Roberts for allowing the execution of these tests and for their remarkable cooperation in every part of the experiment. The efforts of the PhD students from Tufts University especially Dr. Iman Behmanesh and Mr. Amin Nozari during and after the testing period is also appreciated. The Authors also want to thank Ms. Andrea Sacco for her helps during the data post processing. However, the opinions expressed in this paper are those of the authors and do not necessarily represent those of the sponsor or the collaborators.

References 1. Pines, D., Aktan, A.E.: Status of structural health monitoring of long-span bridges in the United States. Prog. Struct. Eng. Mater. 4(4), 372–380 (2002) 2. Doebling, S.W., Farrar, C.R., Prime, M.B.: A summary review of vibration-based damage identification methods. Shock Vib. Dig. 30(2), 91–105 (1998) 3. Sohn, H., Farrar, C.R., Hemez, F.M., Shunk, D.D., Stinemates, D.W., Nadler, B.R., Czarnecki, J.J.: A review of structural health monitoring literature: 1996–2001. Report No. LA-13976-MS, Laboratory LaN, Los Alamos, NM, 2004 4. Brincker, R., Zhang, L., Andersen, P.: Modal identification of output-only systems using frequency domain decomposition. Smart Mater. Struct. 10(3), 441 (2001) 5. Van Overschee, P., De Moor, B.: Subspace Identification for Linear Systems: Theory—Implementation—Applications. Springer Science & Business Media, Berlin (2012) 6. Ventura, C.E., Schuster, N.D.: Structural dynamic properties of a reinforced concrete high-rise building during construction. Can. J. Civ. Eng. 23(4), 950–972 (1996) 7. Gentile, C., Saisi, A.: Ambient vibration testing and condition assessment of the Paderno iron arch bridge (1889). Constr. Build. Mater. 25(9), 3709–3720 (2011) 8. Ivanovi´c, S., Trifunac, M., Novikova, E., Gladkov, A., Todorovska, M.: Ambient vibration tests of a seven-story reinforced concrete building in Van Nuys, California, damaged by the 1994 Northridge earthquake. Soil Dyn. Earthq. Eng. 19(6), 391–411 (2000) 9. Moser, P., Moaveni, B.: Design and deployment of a continuous monitoring system for the Dowling Hall Footbridge. Exp. Tech. 37(1), 15–26 (2013) 10. Reynders, E., Roeck, G.D., Gundes Bakir, P., Sauvage, C.: Damage identification on the Tilff Bridge by vibration monitoring using optical fiber strain sensors. J. Eng. Mech. 133(2), 185–193 (2007) 11. Siringoringo, D.M., Fujino, Y.: System identification of suspension bridge from ambient vibration response. Eng. Struct. 30(2), 462–477 (2008) 12. Astroza, R., Conte, J., Restrepo, J., Ebrahimian, H., Hutchinson, T.: Shake table testing of a full-scale five-story building: system identification of the five-story test structure. In: Proceedings of ASCE Structures Congress, Pittsburgh, PA, 2013 13. Belleri, A., Moaveni, B., Restrepo, J.I.: Damage assessment through structural identification of a three-story large-scale precast concrete structure. Earthq. Eng. Struct. Dyn. 43(1), 61–76 (2014) 14. Moaveni, B., He, X., Conte, J.P., Restrepo, J.I., Panagiotou, M.: System identification study of a 7-story full-scale building slice tested on the UCSD-NEES shake table. J. Struct. Eng. 137(6), 705–717 (2010) 15. Moaveni, B., Stavridis, A., Lombaert, G., Conte, J.P., Shing, P.B.: Finite-element model updating for assessment of progressive damage in a 3-story infilled RC frame. J. Struct. Eng. 139(10), 1665–1674 (2012) 16. Zembaty, Z., Kowalski, M., Pospisil, S.: Dynamic identification of a reinforced concrete frame in progressive states of damage. Eng. Struct. 28(5), 668–681 (2006)

276

S. Yousefianmoghadam et al.

17. Baghaei Naeini, R.: Vibration-based damage assessment and residual capacity estimation of bridges. Ph.D. thesis, University of California, Irvine, CA (2011) 18. Yousefianmoghadam, S., Behmanesh, I., Stavridis, A., Moaveni, B., Nozari, A.: System identification and modeling of a 100-year-old RC warehouse dynamically tested at several damage states. In: Proceedings of 1st International Conference on Natural Hazards and Infrastructure: Protection, Design, Rehabilitation ICONHIC 2016, Chania, Greece, 2016 19. Yousefianmoghadam, S., Behmanesh, I., Stavridis, A., Moaveni, B., Nozari, A., Sacco, A.: System identification and modeling of a dynamically tested and gradually damaged 10-story RC building. Earthq. Eng. Struct. Dyn. (under review) (2017) 20. Welch, P.D.: The use of fast Fourier transform for the estimation of power spectra: a method based on time averaging over short, modified periodograms. IEEE Trans. Audio Electroacoust. 15(2), 70–73 (1967) 21. Oppenheim, A.V., Schafer, R.W., Buck, J.R.: Discrete-Time Signal Processing. Prentice Hall, Englewood Cliffs (1989) 22. Allemang, R.J., Brown, D.L.: A correlation coefficient for modal vector analysis. In: Proceedings of the 1st International Modal Analysis Conference, 1982

Chapter 34

Kronecker Product Formulation for System Identification of Discrete Convolution Filters Lee Mazurek, Michael Harris, and Richard Christenson

Abstract The following paper illustrates a mathematical framework for identifying discrete convolution filters and applies that framework to a time varying dynamic problem. Convolution matrices, formulated using Kronecker products, allow for least means squares solution of arbitrarily structured discrete convolution models. The framework exposes the convolution structure and the filter coefficients independently in order to solve for the time varying weights of a known model, or an unknown time varying filter. Transforms are shown to apply standard discrete spectral processing methods to this architecture. The identification methods were successfully applied to identify the stiffness-like and damping-like operating regimes of a typical viscous damper in a dynamic system subjected to periodic motion. The results show that discrete time varying system identification successfully predicts measured test data and provides intuitive results showing stiffness-like behavior at low force inputs and damping-like behavior for higher force inputs. Keywords Time variant wiener filter • Time frequency analysis • Time variable system identification

34.1 Background The work presented here is an extension of previous system identification methods, which model linear systems using convolution integrals. System theory uses convolution integrals to model the response of a differential equation representing an electrical, mechanical, or other system to an arbitrary input. For continuous linear systems, an output, y, can be calculated given an input, x, and an impulse response filter, h, by Zt h . / x .t  / d

y.t/ D

(34.1)

0

where t is time and £ is the relative time delay between input and output. The continuous impulse response filter can be solved for given an input and output using spectral factorization methods described by Wiener [1] and Kolmogoroff [2]. The discrete form [3] of the above convolution integral is y Œn D

X

 h.m/ x .n  m/

m

for

m  max .n  .p  1/ ; 0/ m  min .n; .r  1//

(34.2)

where lag, m, is used to show the sample delay between input and output, n indicates the output sample, r represents the length of the filter, h, and p represents the length of the input x. The solution for the discrete filter, h, is simplified relative to the continuous filter, and was defined in [3]. The work presented here focuses on the time varying discrete convolution problem y Œn D

X

H .m; n  m/ x .n  m/

(34.3)

m

L. Mazurek • M. Harris • R. Christenson () Civil & Environmental Engineering, University of Connecticut, 261 Glenbrook Road Unit 3037, Storrs, CT, 06269-3037, USA e-mail: [email protected] © The Society for Experimental Mechanics, Inc. 2017 J. Caicedo, S. Pakzad (eds.), Dynamics of Civil Structures, Volume 2, Conference Proceedings of the Society for Experimental Mechanics Series, DOI 10.1007/978-3-319-54777-0_34

277

278

L. Mazurek et al.

where H may be a matrix of time varying impulse response filters, with n  m selecting the impulse response for the corresponding input time. The solution process to find the filter H is the focus of this paper. The structure of time invariant discrete convolution has been well researched. Linear time invariant discrete convolution is known to possess a repeating structure, whose bases can be derived using Kronecker products. Efficient forward convolution solution methods have been implemented using these bases [4, 5]. The solution for discrete time invariant filter, h, can be performed efficiently given knowledge of the repeating structure. Kailath and Sayed [6] discuss this form of matrix structure and provide multiple methods for inversion of these equations. Hansen [7] discusses deconvolution and regularization methodologies for the solution of two dimensional convolution problems related to image processing. These methods are expanded here to address the discrete time variant system identification problem. Statistical methods are well established for the empirical analysis of discrete time convolution systems. Bendat and Piersol [8] provide methods for assessing the quality of fit of single input single output, and multi input multi output system models. Bendat [9] shows how to transform single input single output nonlinear systems in order to employ standard multi input single output linear statistical processing. One objective of the discrete convolution system identification models documented here is that they may be assessed using these standard statistical methods.

34.2 Mathematical Methods Time varying system identification may be performed using convolution matrix products, which are the basis of this paper and are explained here in detail. The convolution sum, as expressed in Eq. (34.3), is rewritten as a matrix vector product by expanding the terms contributing to each output sample and placing zeros elsewhere. The resulting matrix appears as a series of impulse responses with unit sample delays relative to one another. The discrete output vector y is calculated using 2

H00 0 6 6 H H ::: 6 10 01 6 : : 6 : 6 : H11 :: 6 : : y D HŒCV x D 6 6 Hr0 :: :: 6 : 6 6 0 Hr1 :: 6 : :: :: 6 : 4 : : : 0 0

0

3

7 7 7 2 3 7 x0 7 7 6 7 7 6 x1 7 7 : 7 H0p 7 6 7 4 :: 5 7 H1p 7 xp :: 7 7 : 5 Hrp 0 :: :

(34.4)

where the matrix, H[CV] , is equivalent to the time varying convolution operation. Note that for the purpose of this discussion, H[CV] is used to indicate the filter stored in H expanded into a matrix which has convolution structure given by CV (as defined subsequently in Eq. (34.18)). In the above expressions, y is an (n  1) vector, H[CV] is an (n  p) matrix representing a convolution operation, x is a (p  1) vector, and n D p C r  1 where r is the length of each impulse response stored in H. While the above equation is sufficient to solve for output, y, it cannot be used directly to solve for the filter, H, or its convolution structure, CV , given x and y. Generally H[CV] ¤ y x1 because the standard inversion process does not enforce linear dependent structure and x is not directly invertible. The matrices are rearranged using the Kronecker identity as follows y D In HŒCV x

(34.5)

    vec.y/ D xT ˝ In vec HŒCV

(34.6)

where ˝ indicates the Kronecker product, vec( : : : ) indicates the columns of the enclosed matrix stacked as a vector, and In is the identity matrix of size n. The Kronecker product identity above is used to expose the expression vec(H[CV] ), which allows the terms of H[CV] to be grouped and organized into a linear structure as follows   vec HŒCV D CV vec.H/

(34.7)

where CV is the convolution structure and H is a matrix of filter coefficients. The function of the variable CV is to describe the structure of the convolution. The columns of CV represent terms in H and the rows represent output time for given

34 Kronecker Product Formulation for System Identification of Discrete Convolution Filters

279

input time. The above equations are combined into one total expression which contains the convolution structure and the filter independently in the equation as       y D HŒCV x D xT ˝ In vec HŒCV D xT ˝ In CV vec.H/

(34.8)

where vec(y) D y in this equation because it is assumed that y is already a vector. The above expression is a form of the discrete convolution product which can be used to solve for H. The left Moore Penrose pseudoinverse [10], indicated by ( : : : )P is used to solve for H by  P  T y D vec.H/ x ˝ In C

(34.9)

  T   1  T  T x T ˝ In C x T ˝ In C x ˝ In C y D vec.H/

(34.10)

Portions of the above equation are grouped together into equivalent auto and cross-correlation matrices using Kronecker identities as vec.H/ D Rxx 1 Ryx Ryx D

(34.11)

 T  T x ˝ In CV y D CV T .x ˝ In / .1 ˝ y/ Ryx D CV T .x ˝ y/

Rxx D

(34.13)

   T     T  x ˝ In CV xT ˝ In CV D CV T .x ˝ In / xT ˝ In CV Rxx D CV T



(34.12)

  xxT ˝ In CV

(34.14) (34.15)

For the case of linear time invariant convolution (denoted by absence of V subscript on C, but otherwise the same), the convolution structure C is composed of a linear mapping of impulse responses and a shift operator by 2

6 6 6 2 3 6 6 6 S0 Inr 6 S1 I 7 6 6 6 nr 7 6 : 7 6 6 : 7 6 6 : 7 6 CD6 i 7D6 6 S Inr 7 6 6 7 6 6 :: 7 6 4 : 5 6 6 6 Sp Inr 6 6 6 6 4

Ir

 3

7 0p1;r 3 7 7 01;r 7 4 Ir 5 7 7 7 0p2;r 7 7 7 :: 7 : 2 37 7 0i;r 7 4 Ir 5 7 7 7 0p1i;r 7 7 :: 7 7 :

 7 0p1;r 5 Ir 2

2

0 61 6 6 S D 60 6: 4 ::

0 0 1 :: :

0 0 0 :: :

:: :

3 0 07 7  Sij D 1 for i  1 D 1 07 7D Sij D 0 otherwise :: 7 :5

(34.16)

0 0 0 1 0

where the shift operator, S, is an (n  n) matrix which h takes a unit input vector and delays it by a single sample. The structure C can be interpreted as p series of (n  r) matrices mapping impulse h to its appropriately delayed output sample for each input sample. The notation Si is used to indicate the shift matrix raised to the ith power, or multiplied a total of i times, and S0 is an identity matrix. Inr indicates a rectangular identity matrix of dimensions (n  r), where the lower and right columns of the largest square identity matrix are truncated to yield the rectangular identity matrix. The expressions for auto and cross correlation can be shown to the standard expressions in Bendat and Piersol [8] when the time invariant C structure is employed. The solution process for time varying H is similar to the one presented previously, but the variable CV defining the convolution structure now has additional time variance, indicated by the subscript V. The number of terms in H that must be solved for grows significantly relative to the time invariant case and therefore, additional

280

L. Mazurek et al.

constraints on the structure are needed to aid in the solution process. Often, the time variance in a system can be attributed to an underlying state which has a known or measurable relationship to time. If the state is repeated multiple times during an event, the impulse responses associated with the corresponding time can be made linear dependent to reduce the number of overall variables required to be solved. Another method to assess linear dependence over time is to begin with several candidate models and calculate the sample dependent model weight. These methods are a subset of the general time varying convolution structure and are useful to realize overdetermined solutions. For the time varying convolution to be used in this paper, each input sample results in an equal length impulse response. V is a weighting matrix with rows equal to number of unique impulse responses, s, and columns equal to number of input time samples, p. V can be interpreted as an array of time varying weighting vectors, describing each impulse response’s contribution for a given input sample time. In the following notation, a single subscript (vi ) indicates the vector associated with one column, or weights for every candidate filter at a single input time, and the row within that vector (vij ) represents weighting for each candidate filter individually for the given input time. The V matrix is described by 2

v00 6 :: V D 4 : vs0

3 v0p :: 7 D v v : 0 p : 5

(34.17)

vsp

The structure CV uses the terms in V to generate p series of (n  pr) matrices mapping each impulse stored in H to its appropriately delayed output sample for each input sample. The Kronecker product is used to take the original time invariant structure represented by unit shift operations and copy it with a given weighting for each impulse response model, which can be expressed as 3 3 22  v00 Sn 0 Inr v0 T ˝ Sn 0 Inr  6v T ˝ S 1 I 7 66v S 1 I 7 6 6 01 n nr 7 6 1 n nr 7 6 7 7 66 :: :: 6 7 7 66 : : 6 6 6 7 7  7 D 66 CV D 6 T  i 7 6 vi ˝ Sn Inr 7 6 6 v0i Sn i Inr 7 6 7 7 66 :: :: 6 7 7 66 4 4 4 5 5 : : p p T v0p Sn Inr vp ˝ .Sn Inr / 2

33 3 vs0 Sn 0 Inr 66 v S 1 I 77 7 66 s1 n nr 77 7 66 77 7 :: 66 77 7 : 66 77 7 66 77 7 66 vsi Sn i Inr 77 7 66 77 7 :: 66 77 7 44 55 5 : p vsp Sn Inr 22

(34.18)

One difficulty associated with the solution for vec(H) is the number of unknown terms relative to the available knowns. To overcome this difficulty, the matrix V should be constructed to introduce linear dependence. Methods of ensuring full rank include measuring multiple independent records of x and y, and making assumptions related to the amount of variation the impulse response experiences over time. For example, if H is dependent on a repeating state, V can reflect that the same vector hi (representing the ith column of H) appears multiple times. With appropriate structure, and sufficient averaging, it is possible to have more knowns in x and y than unknowns in vec(H). The only difference in the algebraic solution between time varying and time invariant filter, H, is the usage of the known time invariant convolution structure C or the known time variant convolution structure CV . The following discussion describes the solution process for unknown CV if candidate H is known. In other words, if candidate impulse responses are available, the time varying weights of those impulse responses, which best fit a data set, can be solved for.

34.2.1 Solution for Time Varying Weighting Matrix The matrix V describing the time varying weighting of a given column in H at each point in time can be solved for directly. The process involves exchanging between different forms of the convolution product, by   y D xT ˝ In CV vec.H/ D matn .Cv vec.H// x

(34.19)

where the operator matn ( : : : ) is the operator which reverses the vec ( : : : ) operation, and has additional subscript n to ensure the number of rows are unambiguous. The structure CV is given previously and the Kronecker product notation is exchanged back to triple matrix multiply notation below. The following is an excerpt of CV structure which demonstrates the exchange.

34 Kronecker Product Formulation for System Identification of Discrete Convolution Filters

281

 T  i     vi ˝ Sn Inr vec.H/ D vec Sn i Inr H vi

(34.20)

The output y, can now be interpreted as the sum of each element in x multiplied by the appropriately weighted and delayed impulse for its point in time. This representation is shown below in matrix form and as a summation. 2

3 x0 7   0    i     6 6 xi 7 yD Sn Inr H v0 Sn Inr H vi .Sn p Inr / H vp 6 : 7 4 :: 5

(34.21)

xp yD

Xp iD0

 i   Sn Inr H v i xi

(34.22)

Given that each vi is a vector indicating the weighted sum of matrices multiplied by scalar values, their location in the equation can be moved while preserving equivalence by 2

3 v0 7   0    i     6 6 vi 7 yD Sn Inr H x0 Sn Inr H xi .Sn p Inr / H xp 6 : 7 4 :: 5

(34.23)

vp The least means square solution for V is as follows

KD

V D mats .L v/ Q

(34.24)

 1 T vQ D K P y D K T K K y

(34.25)

  0    i     Sn Inr H x0 Sn Inr H xi .Sn p Inr / H xp L

(34.26)

where the vector form of V is 3 v0 6 7 vec.V/ D 4 ::: 5 vp

3 v0 6 7 vec.V/ D L vQ D 4 ::: 5 2

2

or

(34.27)

vp

The variable L has been introduced above to assign linear dependence and improve rank of the pseudo-inverse problem. If L is not set to identity, vQ will contain the unique terms of the weighting vector, which may be projected to a full weighting vector using L. The term KT y is analogous to the cross correlation of H and y weighted by each term in x. The term KT K is analogous to the auto correlation of H weighted by each term in x. For most problems, the solution for time dependent weight of candidate models involves fewer unknowns than the solution for the fully time dependent filter model. Therefore, if candidate models are available, this method can efficiently use limited test data to better understand the time variance of a system.

34.2.2 Equivalence of Time Varying Transfer Function and Time Varying Input The previous equations can be equivalently interpreted as a multi-input single-output relationship where time variance is expressed as a transform on the input term rather than the transfer function. A transform between these two forms can enable the use of conventional statistical tools to understand time varying systems, and can be used to show the equivalent time varying transfer function of linearized systems. Bendat [9] employs a similar method to transform time varying systems in order to apply time invariant single output multi input statistical analysis, however this method is more specifically defined for discrete time varying convolution processes.

282

L. Mazurek et al.

The matrix Z is a linear (kp  p) transform which takes an original x vector and produces a multitude of x vectors (stacked together in vec ( : : : ) format). The following equations show an original time variant convolution structure CV , and the modified multi input time invariant convolution structure C. vec.X/ D .Z x/

(34.28)

    y D xT ˝ In CV vec H

(34.29)

     y D .Z x/T ˝ In C vec.H/ D xT Z T ˝ In C vec.H/

(34.30)

For the case where a linear time invariant multi-input convolution structure, C, and a time dependent transform, Z, are both known, the equivalent time variable convolution structure can be defined uniquely. yD



      xT Z T ˝ In C vec.H/ D xT ˝ In Z T ˝ In C vec.H/   CV D Z T ˝ In C

(34.31) (34.32)

Given CV and C, it is also possible to solve for the least means squares solution for the transform Z.

34.2.3 Spectral Processing Coherence measures and uncertainty bounds may be developed using the time variant to multi input time invariant transform and will be the subject of future studies. For the empirical demonstration in this paper, simplified coherence measures and spectral processing methods were employed. The coherence measures shown in this paper are calculated as the ratio between predicted signal power and the sum of predicted and error power as 2 D

HSxx H  HSxx H  C S""

(34.33)

where HSxx H is the power spectral density of the response predicted by a given filter, and S"" is the power spectral density of the error residual after the prediction is subtracted from the total. The predicted response is calculated using Eq. (34.8) and subtracted from the original response to find the error. HSxx H and S"" are calculated in this study using the Matlab cpsd( : : : ) function with predicted response and response error as inputs respectively. This equation reproduces the standard equations for coherence when the error and the prediction are uncorrelated. This assumption is true for time invariant system identification, and potentially overestimates total signal power and underestimates coherence for the system identification methods shown here. The compromise was selected in order to assure the reader that the identified filters are indeed predictive, while leaving the complexity of multi input coherence to later publications.

34.3 Experimental Methods The effect of low frequency vibration on damping performance in a coupled beam damper system was empirically assessed using the time varying system identification. The experiment was performed under laboratory conditions to verify the identification methodology but is generally application (for example to: dampers, braces and general energy dissipation devices in civil applications). Figure 34.1 shows a notional schematic of the components involved in the laboratory test as well as a picture of the experiment in the structures research lab at the University of Connecticut. The test beam is a 9.1 m long steel pole, with a 0.15  0.15 m cross section and 6.3 mm wall thickness. A Taylor devices damper, model number 1x2 D-TC specified for 111 N force at 3 m/s, was placed at approximately 3.7 m from the base of the beam. The damper was intentionally tested at velocities outside (lower than) the typical operating range of the damper in

34 Kronecker Product Formulation for System Identification of Discrete Convolution Filters

283

Fig. 34.1 Experimental beam damper system. (a) Notional schematic of test setup. (b) Photo of test setup

RMS Tonal Force Level (N)

40 Low-passed white noise (fc=20Hz n=1) Tone (1Hz sine wave) & white noise

30

20

10

0 0

5

10 15 20 25 30 RMS Broadband Force Level (N)

35

40

Fig. 34.2 Broadband and tonal test levels

order to perform identification of stiffening due to time varying seal behavior (the damper is specified at maximum velocity and works as expected at the specified amplitudes). Vibration test hardware was instrumented to the beam. An LDS V408 electrodynamic shaker was placed between the damper and the foundation of the beam at about 2.4 m from the base. A PCB 208C02 piezoelectric load cell was used to measure the shaker input force and a 208C01 load cell was used to measure the reaction force at the damper. A PCB 393C seismic accelerometer (5% bandwidth 0.025–800 Hz) was placed on the beam at the damper attachment location. Time series were recorded using a DataPhysics acquisition system with SignalCalc software. Input was generated using Matlab Simulink with Quanser I/O boards driving the shaker through a consumer grade voltage amplifier. The input signals were designed to assess the linear model using conventional system identification and to introduce steady time variance to assess time varying system identification. Figure 34.2 shows the test events along two axes representing the broadband force level and the tonal force level. The broadband input was used to perform conventional system identification, and to provide a small linear perturbation signal for time varying system identification. The tones were superimposed upon broadband input to repeatably vary the damper between physical regimes. Different levels were used in each axes in order to explore the damper behavior at higher force levels near the specified operating point and at very low levels. The recorded signals of force and acceleration were digitally post processed such that the identified filter would relate force to velocity. This post-processing was performed by digitally filtering the force record with a first order 100 Hz highpass infinite impulse response filter. The result is that in the center of the processing band ( 1, then xt is nonstationary, and its variance grows explosively with time; if j˛j D 1, then the variance of xt will be t 2 , which will grow with time, thus the process is nonstationary. Such a data generating process is termed a unit root process. A unit root process can achieve stationarity by simply differencing it once, so it is also known as a difference stationary process. The test for the presence of unit root processes used in this paper is perhaps the most commonly used statistical test, the augmented Dickey-Fuller (ADF) test, and the steps for implementing the test will only be briefly outlined here, but readers can refer to [10] and [11] for further details. The ADF test is to fit the time series xt to the following form: xt D xt1 C

m X

j xtj C "t

(41.2)

jD1

where  is a differencing operator such that xt D xt  xt1 , the j are the coefficients of the autoregressive terms, m is the lag number. In this regression, a sufficient number of lags should be included to achieve a white noise residual term "t ; an information criteria is a common choice for determining the lag number. In this form, the value of  will determine the stationarity of the series; if  D 0 then xt possesses a unit root and is nonstationary. Therefore, a statistical test is employed with a hypothesis H0 W  D 0 and an alternative hypothesis H1 W  < 0. The test for the null is simply a t test: O D

O se./ O

(41.3)

where O is the least-squares estimate of , and se./ O is the standard error of . O The critical values of the t-statistics are given in [11]. The null hypothesis is rejected if O is smaller that the corresponding critical value, and accepted otherwise. The model form of Eq. (41.2) can also be further adapted to include shift terms and/or trend terms [12]. Having reviewed the fundamentals of unit root processes and their statistical tests, one can now ascertain the nonstationarity of a series through these procedures. It is not difficult to find that the test statistic is the key ingredient in the unit root test, therefore in this paper, the power of the test statistic will be explored, and attempts to measure the degree of stationarity with it, and determine the best possible model form, will be carried out. As previously stated, cointegration is a powerful tool to understand nonstationary data, two or more nonstationary series are cointegrated if a linear combination of them can be found to be stationary. Let xt D .x1t ; x2t ; : : :xmt / denote an m-variate time series, and suppose there exists a vector ˇ that makes ut D ˇ xt

(41.4)

a univariate stationary time series. Here, the vector ˇ D .ˇ1 ; ˇ2 ; : : :ˇm /0 is referred to as a cointegrating vector. Usually there are more than one possible cointegrating relationships for a multivariate series xt , and many methods to estimate

41 An Exploratory Study on Removing Environmental and Operational Effects Using a Regime-Switching Cointegration Method

331

the cointegrating vectors are available in the literature, readers can refer to [13, 14] and [15] for comprehensive reviews. The Johansen procedure, an efficient maximum likelihood (ML) estimator, will be adopted in this paper to estimate the cointegrating relationship in engineering data, successful applications can be found in [4, 6] and [16]. To perform the Johansen procedure, one should have a Vector Error Correction model (VECM) of the m-variate time series xt , which takes the form: Xt D ABT Xt1 C

p1 X

‰j Xtj C ut

(41.5)

jD1

where ut is a m-dimensional vector Gaussian noise series, ut  N.0; /; A and B are two m  r matrices, where r is the rank of the matrix B. Matrix B is the cointegration vector matrix to be found, consisting of r cointegrating vectors. Matrix A is the adjustment matrix. Expression (41.5) is also referred to as the Granger expression theorem, which explicitly depicts the dynamics between the long run equilibrium (cointegration) and short term adjustments. To find the cointegrating matrix B, one needs to employ steps involving decomposition, forming the likelihood function, optimisation and solving a characteristic equation. Finally the estimate of B can be obtained: BO D .ˇ 1 ; ˇ 2 ; : : :; ˇ r /

(41.6)

where ˇ 1 ; ˇ 2 ; : : :; ˇ r are the corresponding cointegrating vectors. As the first cointegrating vector ˇ 1 corresponds to the largest eigenvalue, so it is natural to select ˇ 1 as the “most stationary” cointegrating vector, so as to make the stationary residual series. The Johansen procedure offers an efficient framework that not only estimates multiple cointegrating vectors at the same time, but also produces a test statistic for determining the number of cointegrating vectors. In the SHM context, it is more of interest to estimate the cointegrating vectors than to perform tests on the number of cointegrating vectors, because it is the most stationary combination that one looks for to eliminate the EOV-induced “nonstationary components” in the data. Due to the limited space available here, the details of implementing the Johansen procedure and the full theory behind it will not be covered; again readers interested can refer to [15].

41.3 A Regime-Switching Cointegration Method The motivation behind developing a nonlinear cointegration approach is actually straightforward: operating structures may have significantly different behaviours under different operating conditions. For example, a bridge operating in summer time and the same bridge with icing conditions in winter time will probably have very distinct dynamic responses, linear cointegration may not suffice in this case. Similar phenomena can also be observed in economics. For instance in a case of statistical arbitrage, cointegration is a common technique for choosing a profitable portfolio from the stock market; but once some important monetary policy is announced, the economical impact might have a significant influence on the original cointegrated variables. A new cointegration relationship needs to be evaluated so as to make a new portfolio, the change position here is normally termed a breakpoint. Reviewing the econometric literature on nonlinear cointegration, a mainstay of this is the extension to threshold cointegration, first proposed by Balke and Fomby in 1997 [17]. In their framework, the adjustment term in the cointegrating regression is allowed to shift once some indication variable exceeds a threshold. Furthermore, there are several other variants built on the vector error correction (VEC) model, as expressed above in Eq. (41.5). In [18] and [19] for example, they allow a threshold effect on the lag terms and the intercept term respectively. Gregory and Hansen [20], however, take the opposite direction to let the cointegrating relationship change, or in their terms, shift regime. More specifically, the cointegrating vector can change its value after a certain breakpoint, after which the system will stabilise itself at another long term equilibrium. The position of the breakpoint is unlikely to be determined in advance, thus they calculate the unit root statistic for each possible regime shift, and evaluate the smallest values across all possible breakpoints. As cointegration has proved to be a powerful tool for modelling EOVs, the goal here becomes to further improve it by creating a piecewise linear cointegration model, which can shift its cointegration form with respect to the condition of EOVs. Inspired by Gregory and Hansen’s work, a regime-switching cointegration method will be adopted to address the issue above. Different from their method however, instead of using the Engle-Granger framework, the more efficient Johansen procedure is implemented to estimate cointegrating vectors. The procedure of the method is summarised as follows:

332

H. Shi et al.

1. Choose suitable monitored variables, rearrange the monitored series in the order of environmental or operational variable. 2. Insert a breakpoint at a position ranging from .Œ0:15 N,Œ0:85 N/, where N is the sample size. 3. At each possible breakpoint, the series is split into two halves; use the Johansen procedure to estimate the cointegrating vectors for each half. 4. With the estimated cointegrating vectors, calculate the residual series of both halves and then merge them into one series, and determine the ADF t-statistic of the merged residual series. 5. Repeat procedures from step 2 to 4 at each point from Œ0:15 N to Œ0:85 N, and construct a plot of all ADF statistics with respect to the breakpoint positions. Pick the minimum value of the curve, the corresponding position represents the optimal breakpoint. 6. With the optimal results from step 5, using the environmental or operational variable as an index variable, construct a switching cointegration relationship and a stationary residual series, which should be purged of EOVs and still have the power to predict damage. By utilising the procedure above, one can build a regime-switching cointegration model that is capable of capturing the nonlinear effects of EOVs. In this exploratory study, only the case of two regimes are investigated, systems that accommodate more regimes can be possibly addressed by inserting more breakpoints in the proposed model. Next, the proposed method will be examined with a synthetic example.

41.4 A Case Study To illustrate the proposed method, a simple linear system is simulated, the results from using the conventional cointegration method will be compared. Figure 41.1 shows a four degree-of-freedom (DOF) spring-mass system, where four lumped masses are in a chain with both ends connected to ground. As is known, the stiffness of the metal material changes with temperature, which is a major source of EOVs in SHM data; besides, freezing conditions of structures may always have a great influence on the masses, stiffness and boundary conditions. To simulate the circumstances above, a bilinear relationship between spring stiffness and temperature is created as shown in Eq. (41.7). Real temperature data recorded from the Tamar Bridge is used here as a simulated thermal field [21]. The data length is 10000, and the temperature ranges approximately from 10 ı C to 20 ı C, as plotted in the lower panel of Fig. 41.2. One can also produce nonlinearity by simply letting the third spring behave distinctly from the other springs, as expressed in (41.8). The stiffness-temperature relationship has the following forms:  k1 D k2 D k4 D k5 D  k3 D

0:15  T C 4; if T < 0 0:05  T C 4; if T  0

0:15  T C 5; if T < 0 0:25  T C 5; if T  0

(41.7) (41.8)

Because of the different behaviour of k3 , the nonlinear effect can therefore be introduced into the vibration modes in which the third spring is participating—the second and the fourth modes to be specific. To perform damage detection, a fault in the stiffness of the spring is simulated: after 5000 data points, the stiffness of the second spring k2 becomes 0:8k2 immediately. A small amount of Gaussian noise is added to account for measurement errors. All natural frequencies of the system are identified at every time instant. They are all arranged with respect to time, as shown in the upper panel of Fig. 41.2. Because of the nonlinearity in Eqs. (41.7) and (41.8), one can see that natural frequency series tend to have larger variances in the cold temperature zone (for example, data points around 4000). It is also clear to see that the effect of temperature is significant. The red dashed vertical line in Fig. 41.2 indicates where damage occurs; however, due to the high variance of the series, any information of damage is overwhelmingly masked.

Fig. 41.1 A four-DOF spring mass system

41 An Exploratory Study on Removing Environmental and Operational Effects Using a Regime-Switching Cointegration Method

333

Fig. 41.2 Upper panel: the time series of the four natural frequency series of the system in Fig. 41.1 plotted as a function of time; Lower panel: temperature series plotted against time. Red dashed line imposes damage introduction

Following the conventional cointegration procedures proposed in [4], one can obtain a residual series as shown in Fig. 41.3. Because the underlying cointegration has not been accurately modelled, only the first part of the residual series stays largely stationary, the high variance from the cold zone has been consequently left in the residual, damage indication has therefore failed. By further investigating the mutual relationship of the four natural frequency series in Fig. 41.1, one can see a clear bilinear relationship as plotted in Fig. 41.4. The knee points in the figure correspond to the breakpoint where the cointegrating relationship changes. Thus, a two-regime cointegration model should be appropriate to model this four-DOF system. Following the procedure in Section 3 to validate the performance of the regime-switching cointegration method: 1. Not all data points are needed for estimating the model, only the alternative points from point 5000 to 8000 are chosen as a training set. In order to find the relationship between the frequency series and temperature, the training set is rearranged according to the magnitude of the temperature series, as displayed in Fig. 41.5 along with the temperature series. Denote the shuffled series as ft D .f1t ; f2t ; f3t ; f4t /; t D 1; 2; : : :N, where N is the sample size. Even though the stiffness is set to have a bilinear relationship with temperature, Fig. 41.5 does not show any clear sign of a breakpoint. 2. Next, the ADF test statistic is utilised as a tool to ascertain the position of the breakpoint. As the breakpoint can be anywhere in the series from the beginning to the end, the following step is to evaluate the ADF statistic at every possible breakpoint. However, in practice, it needs a minimum amount of data to calculate the ADF statistic, so only the data in the middle are used to evaluate it, that is the data set in the interval .Œ0:15 N; Œ0:85 N/. Assume a breakpoint is inserted at position , then the original ft .1 W N/ is split into two sets: f1 .1 W /, f2 . C 1 W N/. One then uses the Johansen procedure presented earlier to estimate the cointegrating vector of each set, say ˇ 1 and ˇ 2 , and to construct the residual series at this breakpoint, e D .ˇ 1 f1 I ˇ 2 f2 /, where “;” is used to concatenate these two vector series; the subscript denotes the fact the residual series depends on the position of the breakpoint. 3. Repeat the procedure above to evaluate all the points in .Œ0:15 N; Œ0:85 N/ as breakpoints, and calculate the respective ADF statistics, which are plotted in Fig. 41.6; the horizontal axis represents the number index of the training set. The blank space at the beginning and the end of the figure indicates the fact that ADF statistics are only evaluated in the interval .Œ0:15 N; Œ0:85 N/. The smallest value of the curve is at data point 976, corresponding to the temperature 0.4767 ı C, which is quite close to the simulation assumption.

334

H. Shi et al.

Fig. 41.3 Residual series obtained using the conventional cointegration method, the red dashed line indicates where damage occurs

Fig. 41.4 Mutual relationship between natural frequency series

4. With the estimated best breakpoint and cointegrating vectors correspondingly, one can have the following regimeswitching cointegration relationship which is indexed by the value of temperature:  "t D

147:90  y1t  107:29  y2t  122:96  y3t C 10:69  y4t  3:54; if T  0:4767 4:51  y1t  84:87  y2t  127:87  y3t  165:07  y4t  24:19; if T > 0:4767

(41.9)

Figure 41.7 shows the residual series from (41.9), the red horizontal lines are the three times standard error bars which show the confidence interval of the health state.

41 An Exploratory Study on Removing Environmental and Operational Effects Using a Regime-Switching Cointegration Method

335

Fig. 41.5 Upper panel: natural frequency series rearranged in the order of temperature; Lower panel: temperature series rearranged in the order of magnitude

Fig. 41.6 ADF statistics plot of the training sample points, the lowest point position determines the breakpoint position for the regime switch

336

H. Shi et al.

Fig. 41.7 Residual series of the cointegration model, the vertical red dashed line indicates damage introduction, the two horizontal red lines represent the three standard error bars; the grey shaded areas show where cointegration switches regimes

It is clear that the residual series is stationary before damage introduction, any effect from temperature is effectively eliminated, and the nonlinear behaviour of the frequency response is precisely captured. After 5000 data points, the magnitude of the residual series exceeds the confidence interval immediately, which indicates strongly the occurrence of damage, the overlaid grey areas show where cointegration switches from one regime to the other. The result can be interpreted by the fact that the regime-switching cointegration is estimated with training data under normal condition, and the health state of the system has been accurately modelled. Whenever damage occurs, the long term relationship of the variables no longer holds, thus the residual series turns nonstationary immediately.

41.5 Discussions and Conclusions Despite the fact that the method suggests very good results, one may still argue that shuffling the original series may break the underlying cointegrating relationship, therefore the estimation procedure might be ill-conditioned. This argument is partly true, that rearranging the order of series will surely break the underlying error correction mechanism [as expressed in Eq. (41.5)], but the long term relationship stays the same, or in other words, the rearranged series have the same cointegrating vectors as the original series, because the cointegrating relationships are stacked pointwise in time. One should bear in mind that the final goal here is fundamentally different from the aim of the econometricians, the concern is more about the long term relationship between variables, the short term adjustments are less of interest for the moment. Therefore, it is legitimate to use temperature as a reference series to rearrange the original series, and estimate the cointegrating vectors of the yielded series. This paper is concerned with exploring a new nonlinear cointegration method aiming to address the issue of nonlinear effects of EOVs in SHM data. The proposed method is based on a breakpoint model from econometrics to build a piecewise linear cointegration model. The proposed method is validated with a synthetic case, the results suggest that environmental effects on systems can be successfully removed; it needs the authors to further investigate this approach to solve real engineering problems.

41 An Exploratory Study on Removing Environmental and Operational Effects Using a Regime-Switching Cointegration Method

337

References 1. Farrar, C.R., Worden, K.: Structural Health Monitoring: A Machine Learning Perspective. Wiley, New York (2012) 2. Sohn, H.: Effects of environmental and operational variability on structural health monitoring. Philos. Trans. R. Soc. Lond. A Math. Phys. Eng. Sci. 365(1851), 539–560 (2007) 3. Engle, R.F., Granger, C.W.: Co-integration and error correction: representation, estimation, and testing. Econometrica 55, 251–276 (1987) 4. Cross, E.J., Worden, K., Chen, Q.: Cointegration: a novel approach for the removal of environmental trends in structural health monitoring data. In: Proceedings of the Royal Society of London A: Mathematical, Physical and Engineering Sciences, The Royal Society (2011). doi:p. rspa20110023 5. Cross, E., Manson, G., Worden, K., Pierce, S.: Features for damage detection with insensitivity to environmental and operational variations. In: Proceedings of the Royal Society of London A: Mathematical, Physical and Engineering Sciences, The Royal Society (2012). doi:p. rspa20120031 6. Dao, P.B., Staszewski, W.J.: Cointegration approach for temperature effect compensation in Lamb-wave-based damage detection. Smart Mater. Struct. 22(9), 095002 (2013) 7. Cross, E.J. Worden, K.: Approaches to nonlinear cointegration with a view towards applications in SHM. J. Phys. Conf. Ser. 305, 012069 (2011). IOP Publishing, Bristol 8. Zolna, K., Dao, P.B., Staszewski, W.J., Barszcz, T.: Towards homoscedastic nonlinear cointegration for structural health monitoring. Mech. Syst. Signal Process. 75, 94–108 (2016) 9. Shi, H., Worden, K., Cross, E.J.: A nonlinear cointegration approach with applications to structural health monitoring. J. Phys. Conf. Ser. 744, 012025 (2016). IOP Publishing, Bristol 10. Cross, E.J.: On structural health monitoring in changing environmental and operational conditions. Ph.D. thesis, University of Sheffield (2012) 11. Dickey, D.A., Fuller, W.A.: Distribution of the estimators for autoregressive time series with a unit root. J. Am. Stat. Assoc. 74(366a), 427–431 (1979) 12. Hamilton, J.D.: Time Series Analysis, vol. 2. Princeton University Press, Princeton (1994) 13. Perman, R.: Cointegration: an introduction to the literature. J. Econ. Stud. 18(3), 3–30 (1991) 14. Enders, W.: Applied Econometric Time Series. Wiley, New York (2008) 15. Johansen, S.: Likelihood-Based Inference in Cointegrated Vector Autoregressive Models. Oxford University Press, Oxford (1995) 16. Worden, K., Cross, E., Antoniadou, I., Kyprianou, A.: A multiresolution approach to cointegration for enhanced SHM of structures under varying conditions - an exploratory study. Mech. Syst. Signal Process. 47(1), 243–262 (2014) 17. Balke, N.S., Fomby, T.B.: Threshold cointegration. Int. Econ. Rev. 38, 627–645 (1997) 18. Hansen, B.E., Seo, B.: Testing for two-regime threshold cointegration in vector error-correction models. J. Econ. 110(2), 293–318 (2002) 19. Lo, M.C., Zivot, E.: Threshold cointegration and nonlinear adjustment to the law of one price. Macroecon. Dyn. 5(4), 533–576 (2001) 20. Gregory, A.W., Hansen, B.E.: Residual-based tests for cointegration in models with regime shifts. J. Econ. 70(1), 99–126 (1996) 21. Cross, E.J., Koo, K., Brownjohn, J., Worden, K.: Long-term monitoring and data analysis of the Tamar Bridge. Mech. Syst. Signal Process. 35(1), 16–34 (2013)

Chapter 42

Evaluation of Contemporary Guidelines for Floor Vibration Serviceability Assessment Zandy O. Muhammad, Paul Reynolds, and Emma J. Hudson

Abstract Technological advances in the construction sector and innovative lightweight and large span structural layouts in modern building floors increasingly mean that vibration serviceability is the governing design criterion. As this trend continues, excessive vibrations induced by human activities are becoming a significant concern. Prediction of floor vibrations at the design stage is often done using currently available design guidelines, such as AISC Design Guide 11, Concrete Society Technical Report 43 Appendix G, SCI P354, Concrete Centre CCIP-016 and HiVoSS. In this paper, the aforementioned design guidelines are used to predict the vibration responses of a typical office floor, which are then compared with the actual measured responses. It is clear that different guidelines provide different tolerance limits which make the satisfactory/unsatisfactory decision imprecise. The results show that the case-study floor is unsatisfactory according to CSTR43 App G and CCIP-016, whereas it satisfies the requirements of AISC-DG11, SCI P354 and HiVoSS. Nevertheless, the experimental vibration response indicates that there is a perceptible level of vibrations but with no adverse comments. These discrepancies highlight the need for a better prediction techniques and more reliable assessment criteria. Keywords Vibration serviceability • Floors • Design guidelines • R factor • Pedestrian loading

42.1 Introduction Vibration serviceability of building floors is an area of particular interest in light of advancements in construction technologies and efficient use of materials. Driven by architectural demands for innovative and aesthetically pleasing designs, modern office floors have ever more open-plan layouts and longer spans with fewer internal partitions. In these floors, significant reductions in mass and damping are reported [1, 2] due to the modern paperless offices (computerised layouts) rather than conventional heavy offices (compartmentalised layouts). As a consequence, floors are becoming ever more prone to exhibit excessive vibrations in the range of frequencies generated by human activities, such as walking. A number of design guidelines, available at the design stage, have been developed to predict the vibration performance of floors and their ability to satisfy prescribed serviceability thresholds. These include: • • • • •

American Institute of Steel Construction Design Guide 11 2016 (AISC DG11) [3] Concrete Society Technical Report 43 Appendix G 2005 (CSTR43 App G) [4] Concrete Centre Industry Publication 016 2006 (CCIP-016) [5] European guideline, Human Induced Vibration of Steel Structures 2007 (HiVoSS) [6] Steel Construction Institute publication 354 2009 (SCI P354) [7]

These design codes have provided methodologies to predict the vibration responses of floor systems using multi-mode SDOF approach under single person loading scenario. However, their reliabilities and limitations have not yet been fully investigated, in particular where the floors are on the borderline of being acceptable or unacceptable in terms of vibration response. Amongst the aforementioned guidelines there are different vibration design procedures which vary in both the serviceability assessment and the tolerance limits.

Z.O. Muhammad () • P. Reynolds • E.J. Hudson Vibration Engineering Section, College of Engineering, Mathematics and Physical Sciences, University of Exeter, North Park Road, Exeter EX4 4QF, UK e-mail: [email protected]; [email protected]; [email protected] © The Society for Experimental Mechanics, Inc. 2017 J. Caicedo, S. Pakzad (eds.), Dynamics of Civil Structures, Volume 2, Conference Proceedings of the Society for Experimental Mechanics Series, DOI 10.1007/978-3-319-54777-0_42

339

340

Z.O. Muhammad et al.

This paper examines the application of the above design guidelines for predicting the vibration response of a typical office floor using the design procedures provided by each guideline and evaluates them against actual vibration responses measured under a single person walking. The paper starts with a description of the case study floor, followed by a description of the measurement campaign and FE analysis. Then, the design guidelines with their procedures are employed to predict the vibration response and comparisons are made. The outcome of the analysis is discussed, both in terms of the reliability of the methods to predict accurate response levels and also with respect to the appropriateness of the various tolerance limits.

42.2 Experimental and Analytical Investigation of Case Study Floor 42.2.1 Floor Configuration The floor is a composite steel-concrete construction in a steel framed office building. This floor is irregular by design with the primary beams varying from 7.193 to 10.013 m in length and spanning between column lines, as shown in Fig. 42.1. The secondary beams also range from 9.53 to 13.0 m in length. The floor features composite steel beams supporting slabs cast of light weight concrete poured over steel profiled decking. A total height of 130 mm of concrete slab is used acting compositely with the secondary beams. The majority of the floor area is open plan office space furnished and there are few partition walls. Columns are located along the two sides of the building as well as along the centrelines.

Fig. 42.1 Plan of floor configuration

42 Evaluation of Contemporary Guidelines for Floor Vibration Serviceability Assessment

341

42.2.2 Data Acquisition Experimental Modal Analysis (EMA) was performed to find modal properties (i.e natural frequencies, modal damping and mode shapes) of the floor, which are shown in Fig. 42.2. Full details of the modal testing and vibration monitoring are discussed elsewhere [8] and some key points are repeated here. Four electrodynamic shakers were used to excite the floor and responses were measured using high quality accelerometers (Honeywell QA750). A test gird of 65 test points were utilised for acquisition of frequency response functions (FRF). For walking responses, the accelerometers were located at a point of high response and data were acquired at different pacing rates ranging from 104 steps per minute to 132 steps

Fig. 42.2 First four modes from FE analysis and experimental modal analysis. (a) FE analysis, f1 = 5.23 Hz, m = 36.02 tonnes. (b) EMA f1 = 5.24 Hz,  = 3.16%. (c) FE analysis, f2 = 6.52 Hz, m = 28.54 tonnes. (d) EMA f2 = 6.06 Hz,  = 2.24%. (e) FE analysis, f3 = 6.33 Hz, m = 35.95 tonnes. (f) EMA f3 = 6.58 Hz,  = 1.87%. (g) FE analysis, f4 = 6.87 Hz, m = 39.95 tonnes. (h) EMA f4 = 7.31 Hz,  = 2.60%

342

Z.O. Muhammad et al.

per minute. The walking path was between grid line D-1 and E-6 (Fig. 42.1), since it was noticed that the lowest mode shapes were concentrated in this region and it was within the reach of walking frequency ranges. The response data were sampled at 204 Hz and subsequent to the measurements the following steps were performed to obtain the measured vibration responses: • BS6841 Wb frequency weighting was applied to the acceleration time history, which takes into account the variation of human perception of vibration at different frequencies. • Running root-mean-square (RMS) trends were calculated for the 1 s integration time for the weighted acceleration. • The RMS values for all the weighted accelerations were found. • The peak of running RMS trends was found, which is termed as maximum transient vibration value (MTVV). • Response factor (R-factor) was calculated by dividing the MTVV value by the base curve value of 0.005 m/s2 .

42.2.3 FE Analysis A 3D FE model of the floor structure was developed in ANSYS from the structural drawings. SHELL63 elements were utilised to model the orthotropic composite floor and BEAM188 was assumed to model all the beams and columns. Manual model updating was performed to match the measured frequencies. A modal analysis was conducted to obtain natural frequencies and mode shapes of the floor. There are a significant number of vibration modes less than 12 Hz and due to space limits only the first four mode shapes are shown in Fig. 42.2.

42.3 Vibration Responses Using Current Guidelines This section presents the design procedures available in the vibration guidelines, i.e [3–7] to estimate the vibration responses to a single person walking. The design methodologies of each guideline are briefly discussed, then the results are presented with the corresponding tolerance limits.

42.3.1 Source of Excitation: Walking Loads The walking load model described by each guideline is different and takes various forms. It is widely accepted by the available guidelines that vibration responses of floors are in two types: a resonance build-up for low-frequency floors and a transient response for high-frequency floors. The threshold frequency between these two categories is around 10 Hz. For the considered office floor, the fundamental frequency is less than 10 Hz; hence, the walking load model only relevant to the low-frequency floors is discussed. The walking load model used in AISC DG11, CSTR43 App G, CCIP-016, and SCI P354 is a Fourier series representation considering only the first four harmonics [3–5, 7], the general from is shown in Eq. (42.1). " F.t/ D G 1 C

N X

# ˛n sin.n2fp t C ˆn /

(42.1)

nD1

where, F.t/ D walking load time history (N); G D static weight of a person ( (168 lb) 750 N in AISC DG11; 700 N in both CSTR43 App G, CCIP-016 and 746 N in SCI P354); ˛ D Dynamic Load Factors (DLFs); n D order of harmonic of the walking rate (n D 1, . . . ); fp D pacing frequency (Hz); t D time (sec); ˆ D harmonic phase angle; N D total number of harmonics. The values of DLFs provided by CSTR43 App G and CCIP-016 are statistically defined to have 25% chance of being exceeded, while in AISC DG11 deterministic values are used and in SCI P354 the values depend on pacing frequencies. Also, the pacing frequency ranges between 1.0–2.8 Hz in CSTR43 App G and CCIP-016, whereas the design range covered by SCI P354 and AISC DG 11 is narrowed down to 1.8–2.2 Hz and 1.6–2.2 Hz, respectively.

42 Evaluation of Contemporary Guidelines for Floor Vibration Serviceability Assessment

343

HiVoSS [6, 9], however, assumes a completely different approach by modelling walking as a step-by-step polynomial function with eight terms. The pacing frequency and pedestrian weight is defined probabilistically and it is a cumulative distribution for each combination of the pacing frequency and pedestrian weight.

42.3.2 Dynamic Properties of the Floor by the Guidelines The current guidelines characterise the dynamic properties of the floors by modal parameters, such as natural frequency of the floor, modal mass and damping ratio. Among the design guidelines, CSTR43 App G allows to use the measured modal parameters and it is applicable to all materials of construction. HiVoSS presents slightly a different approach to include transfer function procedure to estimate the modal properties; however, graphs are provided by HiVoSS as a result of beforehand calculation of the transfer function to find the vibration responses. Hence, the graphs can be used directly by reading off the vibration responses. This guideline is only applicable to steel structures, which implies a significant limitation of the use of this guideline. CCIP-016 [5], similar to CSTR43 App G, is applicable to any construction materials, with the extension of being applicable to floors as well as footbridges. Both AISC DG11 and SCI P354 are only applicable to steel structures.

42.3.3 Vibration Response Estimation and Evaluation For low-frequency floors, the resonant response occurs when one of the harmonics of walking matches a frequency of the floor. Mode superposition is an effective tool in all the guidelines to obtain the final response. The guidelines calculate acceleration responses of each mode, then by using the mode superposition the final outcome is obtained. AISC DG11 [3], in particular, suggests using analytical FRFs to determine which mode provides the highest response and thus the peak magnitude of the FRF will be used to estimate the acceleration response. For the considered floor, the peak FRF value obtained from harmonic (steady state) analysis (see Fig. 42.3) is 0.80  103 m/s2 /N, which occurred between grid line B-1 and C-2 (Fig. 42.1).

Fig. 42.3 Peak FRF magnitude from FE harmonic analysis between grid line B-1 & C-2

344

Z.O. Muhammad et al.

Different vibration criteria are provided to evaluate the vibration responses predicted by each guideline. CSTR43 App G and CCIP-016 calculate the response factor (R factor), which is then compared to recommended tolerance limits based on the floor usage. For office floors, similar to the case study floor, the recommend vibration limit is an R factor of 4. However, SCI P354 only provides a higher recommended limit, which is R factor of 8. On the other hand, AISC DG 11 only sets the peak acceleration as the vibration limit, which is 0.5%g for office floors, this value corresponds to an equivalent R factor of 7. HiVoSS considers a different criterion, which is one step root-meansquare (OS-RMS). This value has a dimension of mm/s. It is based on the peak root mean square velocity calculated from the inverse of Fourier transformation of the weighted velocity response [10]. This value provides different “recommended class” as acceptable criteria, which ranges from class A (highly recommend) to class F (not recommended). The OS-RMS for each class is calculated from a combination of walking frequency and pedestrian weights, the 90% percentile of those values are considered to be the highest response under the walking load. In calculating the OS-RMS90 value, the walking path is not taken into account, this implies that the excitation point is kept fixed. Hence, this method is believed to be “semiprobabilistic” [11]. The OS-RMS90 multiplied by a value of 10 gives the equivalent R factor [10]. The recommended values for office floors according to HiVoSS is between a lower limit (5% probability of complaints) of OS-RMS90 D 0.8 mm/s and upper limit (95% probability of complaints) of OS-RMS90 D 3.2 mm/s, which corresponds to an R factor of 32.

42.4 Results and Discussion The vibration serviceability assessment is performed based on the tuned FE model for all the modes less than 12 Hz for CSTR 43 App G, and SCI P354, while 15 Hz for the CCIP-016. The FRF for AISC DG 11 was performed up to frequency of 10 Hz under a unit amplitude load at location of the highest mode amplitude and the response was measured at the same point (Fig. 42.3). The modal damping ratio of 3% was assumed for all modes of vibration. The response was calculated for a range of the floor frequencies to obtain the peak vibration response (i.e R factor). The results of the maximum predicted R factor are shown in Fig. 42.4. It can be seen that all the guidelines predict different values of R factor. CSTR43 App G and CCIP-016 predict an R factor of greater than 4, which results in an unsatisfactory floor evaluation. SCI P354 gives an R factor of 7.86 from the calculations, which results in a positive assessment of the floor. Both AISC DG 11 and HiVoSS predict the equivalent R factor of 4.19 (0.30%g) and 12.75 (OS-RMS90 D 1.275 mm/s), respectively. Hence, the floor is acceptable by AISC and it is within the recommended region by HiVoSS. Also, the distribution of R factor under various pacing frequencies is presented in Fig. 42.5. It is clear that there are R factor values predicted by a wide range of pacing frequencies from CSTR43 App G and CCIP-016, whereas SCI P354 range of pacing frequencies seems to be inadequate.

Fig. 42.4 Predicted R factor by different guidelines with recommended tolerance limits

42 Evaluation of Contemporary Guidelines for Floor Vibration Serviceability Assessment

345

Fig. 42.5 Distribution of different pacing frequency against R-factor in guidelines. (a) Distribution of pacing frequency against R-factor in CSTR43 App G. (b) Distribution of pacing frequency against R-factor in CCIP-016. (c) Distribution of pacing frequency against R-factor in SCI P354

346

Z.O. Muhammad et al.

From the actual response standpoint, the measured R factor is 5.3 (MTVV D 0.026 m/s2 ). This value corresponds “subjectively” to a perceptible level of the vibration by floor occupants, but it resulted in no adverse comments. It is worth noting that the peak values predicted by the guidelines are scattered in comparison to the actual response. The predicted values provide only a single value for prediction with different descriptors (peak acceleration, peak R factor and OS-RMS90 ), which do not give reliable information on the event occurrence. These peak values may not occur as frequently as predicted (i.e the probability of occurrence and exceedance is not known), despite being assessed as unacceptable/acceptable according to the provided recommended limits. These discrepancies highlight that the current design guidelines can potentially result in unreliable assessment of floor vibrations, which may lead to imprecise assessment as satisfactory/unsatisfactory. Therefore, better calculation techniques and more reliable criteria are required to predict more reliably the vibration responses of floors [12].

42.5 Conclusion This paper has highlighted vibration response prediction by contemporary guidelines relative to the measured response obtained from experimental walking measurements. The response prediction is based on a single peak value of acceleration or R factor, which is not a representative value and may not occur as often as expected by the guidances. Also, the response criteria provided by HiVoSS is obviously much higher than its counterparts, despite being on the basis of a probabilistic approach. Whilst different guidelines provide various assessment criteria, the satisfactory and unsatisfactory decision seem to be imprecise. As a result, the serviceability assessment procedure and recommended vibration tolerance limits of different guidelines seem to be unreliable and misleading, since the prediction is based on a single person loading and the single peak value. Hence, various probability of exceedance needs to be defined in order to reflect the actual behaviour of the floor at different excitations. It is clear that there is a need for further research and investigations to carry out extensive work in in-servicing office environments and develop or improve a more reliable assessment tools. Acknowledgements The authors gratefully acknowledge the financial support of the Qatar National Research Fund (QNRF) through grant NPRP8-836-2-353 entitled “A Unified Approach to Vibration Serviceability Assessment of Floors”.

References 1. Hewitt, C.M., Murray, T.M.: Office fit-out and floor vibrations. Mod. Steel Constr. 44, 35–38 (2004) 2. Middleton, C., Brownjohn, J.: Simplified methods for estimating the response of floors to a footfall. In: Structures Congress 2011, Las Vegas, Nevada, vol. 41171, pp. 383–403. ASCE, New York (2011) 3. Murray, T.M., Allen, D.E., Ungar, E.E., Davis, D.B.: Vibrations of steel-framed structural systems due to human activity: AISC DG11, 2nd edn (2016) 4. Pavic, A., Willford, M.: Vibration serviceability of post- tensioned concrete floors. Appendix G, Technical Report 43, 2nd edn. Concrete Society, Slough (2005) 5. Willford, M., Young, P.: A design guide for footfall induced vibration of structures - CCIP-016. The Concrete Centre, Surrey (2006) 6. HiVoSS: Human induced vibrations of steel structures-vibration design of floors (HiVoSS): guideline. European Commission (2007) 7. Smith, A., Hicks, S., Devine, P.: Design of floors for vibration: a new approach(SCI P354). Steel Construction Institute (SCI), Berkshire (2009) 8. Hudson, E.J., Reynolds, P.: Implications of structural design on the effectiveness of active vibration control of floor structures. Struct. Control. Health Monit. 21(5), 685–704 (2014) 9. HiVoSS: Human induced vibrations of steel structures-vibration design of floors (HiVoSS): background document. European Commission (2007) 10. Van Nimmen, K., Gezels, B., De Roeck, G., Van Den Broeck, P.: The effect of modelling uncertainties on the vibration serviceability assessment of floors. In: Proceedings of the 9th International Conference on Structural Dynamics, EURODYN 2014, Porto, pp. 959–966 (2014) 11. Hassan, O.A.B., Girhammar, U.A.: Assessment of footfall-induced vibrations in timber and lightweight composite floors. Int. J. Struct. Stab. Dyn. 13(2), 26 (2013) 12. Reynolds, P., Pavic, A.: Reliability of assessment criteria for office floor vibrations. In: 50th United Kingdom Conference on Human Responses to Vibration, Southampton (2015)

Chapter 43

Excitation Energy Distribution of Measured Walking Forces Atheer F. Hameed and Aleksandar Pavic

Abstract For vibration serviceability of floors, current design guidelines propose different force models to represent human walking on structures. Those models have been derived based on many assumptions to simplify the real force induced by human walking. One of those assumptions states that the force is assumed periodic. Other simplification is that the spectrum of the force is assumed to have very low energy beyond a certain frequency limit, hence it can be neglected in that higher frequency region. Those assumptions have been verified and validated over time for conventional floor structures. However, modern floors are slender, made of lightweight materials, and have strong orthotropic properties and low point stiffness. Hence they feature localized higher modes that could be excited even with small amount of energy. In this paper, real walking forces are used to demonstrate the excitation energy distribution over frequency range of 0–60 Hz. A unique database of 852 vertical continuous ground reaction forces (GRF) measured on an instrumented treadmill due to walking is used for that purpose. Excitation energy is calculated by summing the power of the measured force in the frequency domain. It is found that there are considerable amounts of excitation energy well beyond the frequency limits proposed by the current floor design procedures. Boxplots are presented showing the realistic energy distribution which could excite the higher modes of lightweight and slender floors. Keywords Vibration serviceability • Human walking • Signal energy • Parseval’s theorem • Fourier spectrum

43.1 Introduction While the ultimate limit states used to govern the design of conventional floor structures in the past, serviceability limit states, specifically vibration serviceability, has become the governing design criterion for modern lightweight and slender floor structures. This is because modern floors feature lightweight materials that are strong enough but not stiff enough. Although these floors are designed in accordance with the procedures proposed by the state-of-the-art design guidelines, floors designed “to code” are increasingly failing to meet the required occupants comfort level. This is considered as a serious issue, and suggests that the current design procedure might not be suitable anymore for the design of modern floor structures. To rationalize this problem, it is crucial to identify the following, as proposed by ISO 10137:2007 [1]: 1. Vibration source, 2. Transmission path, and 3. Vibration receiver. This paper investigates the vibration source, which could be any human activity that can excite a structure such as walking, jogging, or running. For floor structures, walking is the most common human activity that could excite the structure. Current design guidelines propose different forcing models that can be used to represent excitation due to human walking on a

A.F. Hameed () Vibration Engineering Section, College of Engineering, Mathematics, and Physical Sciences, University of Exeter, Kay Building, North Park Road, Exeter, EX4 4QF, UK Civil Engineering Department, University of Anbar, Ramadi, Anbar, 31001, Iraq e-mail: [email protected] A. Pavic Vibration Engineering Section, College of Engineering, Mathematics, and Physical Sciences, University of Exeter, Kay Building, North Park Road, Exeter, EX4 4QF, UK Full Scale Dynamics Ltd., 40 Leavygreave Road, Sheffield, South Yorkshire, S3 7RD, UK e-mail: [email protected] © The Society for Experimental Mechanics, Inc. 2017 J. Caicedo, S. Pakzad (eds.), Dynamics of Civil Structures, Volume 2, Conference Proceedings of the Society for Experimental Mechanics Series, DOI 10.1007/978-3-319-54777-0_43

347

348

A.F. Hameed and A. Pavic

structure. Although those forcing models are easy to implement, they are based on many assumptions, hence are not able to reflect real walking scenarios. All these models are assuming periodicity of human walking as opposed to real walking which are considered narrow-band random forces. Moreover, it is assumed that human walking forces have negligible ability to excite higher modes of vibration with natural frequencies above certain value. For example, when considering resonant vibration analysis, the proposed forcing model has only four harmonics, which extend up to 11.2 Hz [2, 3]. In this paper, measured walking forces are used to demonstrate how the walking excitation energy is actually spread over a much wider range of frequencies that goes well beyond the suggested level by the current design guidelines.

43.2 Methodology The measured walking forces were collected from 85 participants using an instrumented treadmill where 852 continuous vertical walking force time histories were recorded [4]. The walking forces were recorded with a sampling frequency of 200 Hz and the speed of each test subject was controlled, hence each measured force has a unique pacing frequency. The test sequence can be cited in details by Brownjohn et al. [4]. A typical normalized (by the weight of the test subject) walking force time history is shown in Fig. 43.1 along with its discrete Fourier amplitudes for a data block of 20.48 s. To determine the pacing frequency of each measured walking force, the Fourier spectrum was analyzed and all forces were categorized into groups according to their pacing frequencies. The distribution of all measured forces according to their pacing frequency is shown in the histogram in Fig. 43.2. The total energy, E, of a continuous signal x(t) is defined as [5]: Z1 ED

x2 .t/dt

(43.1)

1

In the case of a discrete-time signal, the total energy is defined by [5]: ED

1 X

x2 Œn

nD1

Fig. 43.1 Typical walking force time history and Fourier amplitudes (Pacing frequency is 2 Hz)

(43.2)

43 Excitation Energy Distribution of Measured Walking Forces

349

Fig. 43.2 Histogram of pacing frequencies

Where x[n] is the discrete-time signal, and n is the sample number. The result of Eq. (43.2) is the energy of the whole signal. However, when the energy of a desired frequency band is required then the calculation is best performed in the frequency domain using Parseval’s theorem. Parseval’s theorem is given as [6]: ED

1 X nD1

x2 Œn D

1 1 X jX.k/j2 N kD1

(43.3)

Where X(k) is the Discrete Fourier Transform of the discrete-time signal x[n] and k is the sample number in the frequency domain. Parseval’s theorem could be interpreted as follows: while the energy of the total signal could be calculated by summing the squared value of each sample in the time domain, it is also possible to get the same results by summing the squared spectral lines in the frequency domain. In this paper, Eq. (43.3) is used to evaluate the energy of each walking force time history over predefined frequency range. It is important to emphasize that the energy calculated using any of the Eqs. (43.1)–(43.3) is a signal metric and not the energy defined in physics as “the ability to do work”. The physical energy has units of Joules, while the ‘signal energy’ has units that depend on the units of the signal itself: (unit2 /s) or (unit2 /Hz).

43.3 Results As explained in Sect. 43.2 Methodology, the forces were divided into groups according to their corresponding pacing frequencies. This allowed forces of similar frequency characteristics to be processed independently. The process was based on the analysis of the Fourier amplitudes of each force. The Fourier amplitudes of all forces with pacing frequency of 2 Hz, for example, is shown in Fig. 43.3. Using Eq. (43.3), the signal energy of each force was calculated in frequency bands of 1 Hz for frequency range between 0 and 60 Hz. The 1 Hz frequency band is centered around an integer frequency. For instance, the signal energy at 2 Hz is calculated for a frequency band between 1.5 and 2.5 Hz. The excitation energy is presented statistically by means of box plots. Figure 43.4 shows the calculated energies for the group with pacing frequency of 2 Hz shown previously in Fig. 43.3. It can be seen that the excitation energy is noticeable beyond the limit of 11.2 Hz. The red crosses of the box plot shown in Fig. 43.4 are the outliers [7]. For better representation, Fig. 43.5 shows the total of signal energy over specified frequency bands defined as: 1. 0.5–2.8 Hz: this is where the first harmonic of walking occurs. 2. 2.9–11.8 Hz: this is where the second, third, and fourth harmonics appear.

350

A.F. Hameed and A. Pavic

Fig. 43.3 Fourier amplitudes of 48 walking forces with pacing frequency of 2.00 Hz

Fig. 43.4 Energy distribution of 48 walking forces with pacing frequency of 2.00 Hz

3. 11.9–30 Hz: this is where other harmonics are visually recognized (from Fig. 43.4). 4. 30.1–60 Hz: this is the rest of the frequency range of interest. The first two bands, i.e. up to 11.2 Hz, are the only part of the walking force that is been considered by the current design procedures. It can be noticed from Fig. 43.5 that the other parts of walking forces, i.e. beyond 11.2 Hz, cannot be neglected if the excited structure has modes of vibration in that range of frequencies. Especially when those modes are localized and generate frequency response function (FRF) amplitudes higher an order of magnitude than the FRFs corresponding to the lower modes of vibration.

43 Excitation Energy Distribution of Measured Walking Forces

351

Fig. 43.5 Accumulative energy over certain frequency bands of 48 walking forces with pacing frequency of 2.00 Hz

43.4 Conclusions Analysis of the excitation energy was conducted using 850 measured walking forces from 85 participants. The analysis showed that the energy is distributed well beyond the limits of the proposed walking models of the current design guidelines. Boxplots were used to statistically represent the results where a pacing frequency of 2 Hz was considered. It is suggested that these higher frequency ranges should not be neglected especially when calculating the vibration response of structures with significant contribution from higher modes of vibration. Acknowledgements The database of walking forces was created courtesy of funding by the UK Engineering and Physical Sciences Research Council, Grant EP/E018734/1 (Human walking and running forces: novel experimental characterization and application in civil engineering dynamics). The paper was prepared with the support of the Engineering and Physical Sciences Research Council (EPSRC) grant reference EP/G061130/1 (Dynamic Performance of Large Civil Engineering Structures: An Integrated Approach to Management, Design and Assessment) for which the writers are grateful. The financial support of The Higher Committee for Education Development in Iraq (HCED IRAQ scholarship reference GD-13-5) is highly appreciated as well.

References 1. ISO 10137:2007: Bases for Design of Structures – Serviceability of Buildings and Walkways Against Vibrations – Reviewed in 2012, 2nd edn. Internation Standards Organisation, Geneva (2012) 2. Willford, M.R., Young, P.: A Design Guide for Footfall Induced Vibration of Structures – CCIP-016. The Concrete Centre, Slough (2006) 3. Pavic, A., Willford, M.R.: Vibration Serviceability of Post-tensioned Concrete Floors – CSTR43 App G, Append. G Post-Tensioned Concr. Floors Des. Handb. – Tech. Rep. 43, pp. 99–107 (2005) 4. Brownjohn, J.M.W., Racic, V., Chen, J.: Universal response spectrum procedure for predicting walking-induced floor vibration. Mech. Syst. Signal Process. 1–15 (2015). doi:10.1016/j.ymssp.2015.09.010 5. Haykin, S., Van Veen, B.: Signals and Systems, 1st edn. Wiley, Hoboken (1998) 6. Smith, J.O.: Mathematics of the Discrete Fourier Transform (DFT) with Audio Applications, 2nd edn. W3K Publishing (2008) 7. Montgomery, D.C., Runger, G.C.: Applied Statistics and Probability for Engineers, 3rd edn. Wiley, Hoboken (2002)

Chapter 44

Identification of Human-Induced Loading Using a Joint Input-State Estimation Algorithm Katrien Van Nimmen, Kristof Maes, Peter Van den Broeck, and Geert Lombaert

Abstract This paper uses a state-of-the-art joint input-state estimation algorithm to identify the modal load induced by a single pedestrian on a laboratory structure. The experimental setup involves a simply-supported concrete slab with a length of 7 m. A dynamic model of the lab structure is constructed from a finite element model that is calibrated using a set of experimentally identified modal characteristics. The estimated modal load as is compared with the numerically predicted modal load which uses the average single-step walking load as determined from direct force measurements, the location of the individual steps as identified from video processing and a numerical model of the structure. For the time interval where the pedestrian is crossing the slab, the estimated modal load is found to be in good agreement with the numerically predicted values. Following the last footstep of the pedestrian, the slab passes into a decaying free vibration whereby an exponentially decaying estimated input compensates for small errors in the modal properties. Keywords Human-induced vibrations • Force identification • Footbridge • Human-structure interaction • Vibration serviceability

44.1 Introduction Although the dynamic performance of footbridges under high crowd densities is often imperative for design, the available load models are rudimentary and have virtually never been verified [1]. Concerns about these load models are further strengthened by the fact that human-structure interaction (HSI) phenomena are not well-understood [2]. In-field observations are the only way to obtain detailed and accurate information on representative operational loading data [1, 3]. As direct force measurements are in this case practically infeasible, inverse force identification where the input forces are reconstructed from the resulting vibrational response and a dynamic model of the structure, constitutes a promising alternative. This contribution uses a joint input-state estimation technique as described in Maes et al. [4] to identify the modal load induced by a single pedestrian on a laboratory structure. The results are compared to the numerically predicted modal load which is based on the average single-step walking load as determined from direct force measurements, the location of the individual steps as identified from video processing and a numerical model of the structure.

44.2 Experimental Setup The laboratory structure consists of a simply-supported hollow-core pre-stressed concrete slab with a span of 7 m (Fig. 44.1). The structure only has a single mode, the vertical bending mode, with a natural frequency within the dominant spectrum of pedestrian excitation (< 10 Hz). The corresponding modal parameters (fQj D 6:05 Hz, Qj D 0:39 Hz) are experimentally identified [5].

K. Van Nimmen () • P. Van den Broeck Department of Civil Engineering, Structural Mechanics, KU Leuven, B-3001 Leuven, Belgium Department of Civil Engineering, Technology Cluster Construction, Structural Mechanics and Building Materials Section, KU Leuven, Technology Campus Ghent, Ghent, Belgium e-mail: [email protected] K. Maes • G. Lombaert Department of Civil Engineering, Structural Mechanics, KU Leuven, B-3001 Leuven, Belgium © The Society for Experimental Mechanics, Inc. 2017 J. Caicedo, S. Pakzad (eds.), Dynamics of Civil Structures, Volume 2, Conference Proceedings of the Society for Experimental Mechanics Series, DOI 10.1007/978-3-319-54777-0_44

353

354

K. Van Nimmen et al.

Fig. 44.1 The hollow-core pre-stressed concrete slab excited by a single pedestrian

The experimental study considered a single pedestrian walking along the slab (see Fig. 44.1). The vertical acceleration response of the slab was recorded at 13 locations, uniformly distributed along its length. The pedestrian is instrumented with sensors allowing a 3D tracking of the motion. The registered pedestrian motion allows to identify the average step frequency and the onset of each step [6]. The location of the individual footsteps are identified from video processing (Fig. 44.1). The results show that the average step frequency of the pedestrian equals fQs D 1:98 Hz.

44.3 Results The forces are estimated using a data set consisting of one displacement (nd;d D 1) and 13 acceleration measurements (nd;a D 13. The noise covariance matrices Q and S used in the force identification are assumed to be zero as the force pŒk accounts for all excitation present. The matrix R in this case accounts for sensor noise and is constructed from the (known) standard deviation of the measurement noise [4]. The initial state estimate vector xŒ0j1 and its error covariance matrix PŒ0j1 are both assumed zero. To verify the results of the joint input-state estimation algorithm, the estimated forces are ideally compared to the directly measured input force. Although direct measurements of the pedestrian load are not available in this case, a reasonable approximation of the real walking load is obtained using a generalized single-step load model characterized by the weight of the person and the identified pacing rate [6]. The average step frequency (fQs D 1:98 Hz) and the onset of each step are identified from the tracked pedestrian motion [6]. The single-step walking load follows from previous research involving the same participant, whereby the ground reaction forces (GRFs) were measured directly by an instrumented split-belt treadmill [6]. The averaged vertical single-step walking load for a step frequency of 2.00 Hz was determined from more than 200 consecutive steps. The variations in amplitude of the single-step walking load resulting from small variations of the pacing rate (e.g. 1:98 < fs < 2:00 Hz) are found to be negligible [7]. Moreover, the resulting forces and structural response are found to be much more sensitive to the timing of successive footfalls than to small variations in force amplitude or contact time of the single-step walking load [8]. Together, the averaged vertical single-step walking load (ps ) and the identified onset of each step in time (tq ), allow to reconstruct the force pq due to step q in time:   pq .t/ D  t  tq ps t  tq ; 



( with  .t/ D

1

0  t  tc

0

otherwise

(44.1)

with t [s] the general time of the experiment and tc the duration of contact between the foot and the supporting structure. Taking into account the identified location of each step and the mass-normalized mode shape, the modal load induced by the pedestrian is calculated. Figure 44.2 compares the numerically predicted modal load and individual foot traces with the results of the joint inputstate estimation algorithm. Both the comparison in time (Fig. 44.2a) and frequency domain (Fig. 44.2b) show a fairly good agreement between the numerical predictions and the vibration-based estimated input. The harmonics of the walking load can be clearly observed in the corresponding amplitude spectrum (see Fig. 44.2b).

44 Identification of Human-Induced Loading Using a Joint Input-State Estimation Algorithm

(a)

355

(b) 30 Modal force

Modal force

0 −20 −40

20 10

peri 4

6 Time [s]

Force [N]

(c)

8

10

0

0

5

10 15 Frequency [Hz]

20

800 600 400 200 0

4

4.5

5

5.5 6 Time [s]

6.5

7

7.5

Fig. 44.2 Comparison between the identified (green) and simulated (blue) modal load corresponding to fundamental bending mode of the slab: (a) time series and (b) amplitude spectrum of the peri phase (when the pedestrian is on the slab) and (c) the reconstructed foot traces

44.4 Conclusions This contribution uses a state-of-the-art inverse force identification technique to reconstruct the pedestrian-induced load from the resulting vibrational response and a dynamic model of the structure. The experimental study involves a simply-supported concrete slab with a length of 7 m excited by a single pedestrian. The results of the joint input-state estimation algorithm are verified by comparison with the numerically predicted modal load. A good agreement is found between the estimated modal load and the corresponding numerical predictions. Acknowledgements The first author is a post-doctoral fellow of the Research Foundation Flanders (FWO). The financial support is gratefully acknowledged.

References 1. Georgakis, C., Ingólfsson, E.: Recent advances in our understanding of vertical and lateral footbridge vibrations. In: Proceedings of the 5th International Footbridge Conference, London, July (2014) 2. Bruno, L., Venuti, F.: Crowd-structure interaction in footbridges: modelling, application to real case-study and sensitivity analysis. J. Sound Vib. 323, 475–493 (2009) 3. Živanovi´c, S.: Benchmark footbridge for vibration serviceability assessment under vertical component of pedestrian load. J. Struct. Eng. 138, 1193–1202 (2012) 4. Maes, K., Van Nimmen, K., Lourens, E., Rezayat, A., Guillaume, P., De Roeck, G., Lombaert, G.: Verification of joint input-state estimation for force identification by means of in situ measurements on a footbridge. Mech. Syst. Signal Process. 75, 245–260 (2016) 5. Van Nimmen, K., Maes, K., Van den Broeck, P., De Roeck, G., Lombaert, G.: Inverse identification of pedestrian-induced loads. In: Sas, P., Moens, D., Denayer, H. (eds.) Proceedings of ISMA 2016 International Conference on Noise and Vibration Engineering, Leuven, Sept (2016) 6. Van Nimmen, K., Lombaert, G., Jonkers, I., De Roeck, G., Van den Broeck, P.: Characterisation of walking loads by 3D inertial motion tracking. J. Sound Vib. 333, 5212–5226 (2014) 7. Raci´c, V., Pavi´c, A., Reynolds, P.: Experimental identification and analytical modelling of walking forces: a literature review. J. Sound Vib. 326, 1–49 (2009) 8. Middleton, C.: Dynamic performance of high frequency floors. PhD thesis, University of Sheffield (2009)